From 6c427b5e7def9cb83758a649a99ebae1425a9bc1 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=EC=B5=9C=EC=98=81=EA=B8=B0?= <dch0202@rsquare.co.kr>
Date: Fri, 31 Jul 2026 11:28:14 +0900
Subject: [PATCH 1/3] knowledge: ingest 3 verified insight(s)

---
 .dev-loop/INGEST_REPORT.md                    | 146 +++++++++++++++---
 INDEX.md                                      |   4 +-
 log.md                                        |   3 +
 .../integrations/externally-owned-defaults.md |  70 +++++++++
 .../integrations/llm-response-completeness.md |  66 ++++++++
 wiki/backend/index.md                         |   7 +
 .../config/environment-config.md              |   5 +-
 wiki/platforms/index.md                       |   1 +
 .../processes/background-services.md          |   2 +-
 .../non-interactive-cli-invocation.md         |  71 +++++++++
 10 files changed, 348 insertions(+), 27 deletions(-)
 create mode 100644 wiki/backend/common/integrations/externally-owned-defaults.md
 create mode 100644 wiki/backend/common/integrations/llm-response-completeness.md
 create mode 100644 wiki/platforms/processes/non-interactive-cli-invocation.md

diff --git a/.dev-loop/INGEST_REPORT.md b/.dev-loop/INGEST_REPORT.md
index 2ff3871..c6f4745 100644
--- a/.dev-loop/INGEST_REPORT.md
+++ b/.dev-loop/INGEST_REPORT.md
@@ -1,35 +1,137 @@
-# Knowledge flush — 1 insight
+# Knowledge flush — 3 insight(s)
 
-Source: RNR-3440 (사내 잠재매물 주간 추출 스크립트 메모리 피크 저감). Candidate:
-"QueryPie 프록시 경유로 대용량 결과를 스트리밍할 때 server-side named cursor 대신
-일반 커서 + `fetchmany` + openpyxl `write_only`."
+Drained from `~/.dev-loop/queue` (3 pending candidates, harvested 2026-07-31 from
+sessions in `~/.claude/tmp` and `meetingSummary`). Result: **3 new pages** (one of them
+opening a new `backend/common/integrations` category) plus one merge-edit into an
+existing infrastructure page. No auto-merge.
 
 ## Verified best-practice
 
-**Claim 1 — psycopg2 server-side (named) cursor requires a transaction; fails under autocommit.**
-- Source: psycopg2 usage docs — `https://github.com/psycopg/psycopg2/blob/master/doc/src/usage.rst` (via Context7). Quote: "Named cursors are typically created 'WITHOUT HOLD', meaning they exist only within the current transaction. Attempting to fetch from them after a commit or in autocommit mode raises an exception."
-- Matches my live repro (`can't use a named cursor outside of transactions`). → **verified**
+### 1. A non-interactive flag does not stop a CLI reading stdin
+**Claim.** When invoking a prompt-capable CLI from a script/CI/agent harness, pass the
+non-interactive flag *and* detach fd 0 (`</dev/null` or `nohup`); if it still hangs,
+split client-vs-server by whether the request appears in the server's access log.
 
-**Claim 2 — a client-side (default) cursor pulls the whole result set to the client on execute; `fetchmany` only caps the Python-list explosion.**
-- Source: psycopg2 cursor/usage docs + FAQ (named-cursor advantage = "data is fetched in chunks … minimal client memory"). By contrast the default cursor buffers the full result in libpq. → **verified**
+Sources checked (all fetched live, 2026-07-31):
+- [GNU coreutils `nohup`](https://www.gnu.org/software/coreutils/manual/html_node/nohup-invocation.html)
+  — "If standard input is a terminal, redirect it so that terminal sessions do not
+  mistakenly consider the terminal to be used by the command. Make the substitute file
+  descriptor unreadable, so that commands that mistakenly attempt to read from standard
+  input can report an error." Documents this as a GNU extension with the portable form
+  `nohup command 0>/dev/null` — which is what makes the page's BSD edge case correct.
+- [OpenBSD `ssh(1)`](https://man.openbsd.org/ssh) — `-n` "Redirects stdin from /dev/null
+  (actually, prevents reading from stdin). This must be used when ssh is run in the
+  background." Primary evidence that mature CLIs ship an explicit stdin-detach flag
+  *because* background invocation otherwise blocks.
+- [`git` docs, `GIT_TERMINAL_PROMPT`](https://git-scm.com/docs/git) — "If this Boolean
+  environment variable is set to false, git will not prompt on the terminal", i.e.
+  prompting on the terminal is the default. Second independent instance of the pattern.
 
-**Claim 3 — openpyxl `write_only` gives near-constant memory (<10 MB); one save only; lxml is for serialization speed, not the memory saving.**
-- Source: openpyxl Optimised Modes — `https://openpyxl.readthedocs.io/en/stable/optimized.html` (via WebSearch). "keeping memory usage under 10Mb"; "A write-only workbook can only be saved once"; "make sure you have lxml installed" for large dumps (speed).
-- This **corrects** the raw candidate's "lxml unnecessary" → precise form: unnecessary *for the memory win*, recommended *for large-dump speed*. Confirmed by my server test (write-only worked with lxml absent). → **verified**
+Reproduction from the session: `pi 0.79.1 --tools` in the foreground hung twice
+(300s, 150s; 0 bytes written), while the LiteLLM gateway log showed **zero requests from
+that IP in the window** — so the fault was local, not the model. The identical command
+under `nohup` (stdin `/dev/null`) completed a tool call immediately.
 
-**Claim 4 — QueryPie blocks `BEGIN`, so server-side cursor is impossible there.**
-- Environment-specific, no external source. Live repro in gui context: `autocommit=False` + named cursor → `[ENGINE] No permission to execute BEGIN statement`. → **field-tested**. Generalized in the page to "a read-only access-control proxy that blocks transaction control", with QueryPie as the concrete example (not a product-specific page).
-- Memory figure 838 MB → 38 MB (300k synthetic rows) is my RNR-3440 measurement (`ru_maxrss`, separate processes).
+**confidence: verified** — mechanism documented by three primary sources, plus a local
+reproduction.
+
+### 2. HTTP 200 from `/chat/completions` is not a complete answer
+**Claim.** Before using LLM output as an artifact, fail on `finish_reason == "length"`
+and on blank `content`; when `content` is blank and a reasoning field is populated,
+report that a reasoning model spent the output budget rather than publishing the
+reasoning text.
+
+Sources checked (all fetched live, 2026-07-31):
+- [OpenAI reasoning guide](https://developers.openai.com/api/docs/guides/reasoning) —
+  reasoning tokens "still occupy space in the model's context window and are billed as
+  output tokens"; truncation "might occur before any visible output tokens are produced,
+  meaning you could incur costs for input and reasoning tokens without receiving a
+  visible response"; detect via `status == incomplete` / `max_output_tokens`; reserve
+  ≥25,000 tokens while calibrating. This is exactly the failure mode the candidate claimed.
+- [OpenAI chat-completion object](https://developers.openai.com/api/docs/api-reference/chat/object)
+  — `finish_reason: "length"` = "the maximum number of tokens specified in the request
+  was reached"; `tool_calls` and `content_filter` are the other non-`stop` values (the
+  `tool_calls` case is why the page carves out legitimately-blank content).
+- [vLLM reasoning outputs](https://docs.vllm.ai/en/latest/features/reasoning_outputs.html)
+  — reasoning goes in `reasoning` (renamed from `reasoning_content`), final answer in
+  `content`.
+- [LiteLLM reasoning content](https://docs.litellm.ai/docs/reasoning_content) — LiteLLM
+  normalizes to `message.reasoning_content` (+ `thinking_blocks` for Anthropic).
+
+The two gateways disagreeing on the field name is a fact the candidate did not contain;
+it became an edge-case row ("check both keys before concluding no reasoning").
+
+Reproduction from the session (2026-07-28, LiteLLM on dgx): alias `chat-default` with a
+9,317-token prompt returned **HTTP 200, `finish_reason=length`, `content` 0 chars,
+`reasoning_content` 8,173 chars**; the same request to the non-reasoning `coder-fast`
+returned `finish=stop` with 2,418 complete chars.
+
+**confidence: verified** — the mechanism and every field name are documented; the
+measurement matches.
+
+### 3. Defaults naming repo-external resources need a re-check at merge time
+**Claim.** When a default names a resource the repo does not own, re-query the owner's
+catalog at review/merge time and resolve the name at startup.
+
+Sources checked (all fetched live, 2026-07-31):
+- [OpenAI deprecations](https://developers.openai.com/api/docs/deprecations) — after a
+  shutdown date a model "will no longer be accessible"; notice is ≥6 months (GA),
+  ≥3 months (specialized variants), and **as little as 2 weeks for preview models**. That
+  last number is what makes "verified in the PR body" a perishable claim, and it is now
+  the page's rule for tracking shutdown dates.
+- [`GET /v1/models`](https://developers.openai.com/api/docs/api-reference/models/list) —
+  "Lists the currently available models, and provides basic information about each one
+  such as the owner and availability." Confirms the review-time check is a real,
+  documented endpoint.
+- [LiteLLM proxy model discovery](https://docs.litellm.ai/docs/proxy/model_discovery) —
+  the proxy's own `/v1/models` reports what is actually available *behind that gateway*
+  (`check_provider_endpoint: true` for wildcards). This corrected the candidate: the
+  catalog to query is the gateway's, not the upstream provider's.
+
+Field evidence: PR #15 review (2026-07-28) — default summary alias `dgxb/kanana` had
+disappeared from the LiteLLM catalog and returned 400, while the PR's own end-to-end
+measurement (7/12, 857 chars in 11s) had been true when written; code, tests, and build
+all still passed.
+
+**confidence: verified** — the retirement mechanism and both catalog endpoints are
+documented; the incident supplies the review-timing lesson.
 
 ## Existing-layer check
 
-- Pages read: `databases/index.md`, `databases/query-optimization/keyset-pagination.md`, `backend/python/index.md`.
-- Overlap: keyset-pagination is the nearest neighbor (both handle large result sets) but a **distinct** topic — pagination splits the read into many bounded queries; this page streams a *single* query's result in chunks. Not a duplicate → new page + **bidirectional `related` link** added to both.
-- backend/python has no DB-cursor page; the psycopg2/openpyxl specifics live as concrete examples inside the databases page rather than a separate python page (no duplication).
-- Conflicts: none found.
+Pages read in full before writing anything:
+
+| Read | Outcome |
+|------|---------|
+| `INDEX.md`, `AGENTS.md`, `templates/page.md`, `skills/wiki-ingest/SKILL.md` | Routing protocol + page format applied (positive guidance, `Instead of` pairing, ≤120 body lines, sourced frontmatter) |
+| `wiki/platforms/index.md` + `processes/background-services.md` | **Adjacent, not duplicate.** background-services owns *lifetime after the session ends* (nohup vs launchd vs systemd, KeepAlive, minimal environment). Insight 1 is about *the invocation blocking on fd 0*. Its one genuine overlap — "harness background task killed at session end" — already lives in background-services, so the new page links there instead of restating it. `related:` added **both ways** |
+| `wiki/platforms/shells/portable-shell-scripts.md`, `environment/path-resolution.md`, `toolchains/version-management.md`, `tools/bsd-vs-gnu-cli.md` | Grepped all of `wiki/platforms` + `wiki/debugging` for `stdin\|/dev/null\|tty\|interactive`: every hit is about *non-interactive shells not loading rc files* (PATH/shims), never about a blocked stdin read. Confirmed gap. Linked to bsd-vs-gnu-cli for the macOS `timeout` gap, and to path-resolution in an edge case |
+| `wiki/debugging/index.md` (+ methodology/signals rows) | Considered as the home for insight 1's client-vs-server split. Kept in platforms because the artifact being changed is the *invocation*; the diagnostic step is one table in that page and links to `debugging-methodology-hypothesis-testing` rather than duplicating it |
+| `wiki/backend/index.md` and every `backend/common/**` load-when line | No page covers consuming a third-party API's *response semantics*. `api-design/*` governs the API you publish; `reliability/timeouts-and-retries` governs transport (its "never retry 400" row is why the new page frames a `length` retry as a *new* request — cross-linked). Confirmed gap |
+| `backend/node/boundaries/runtime-validation.md`, `backend/python/boundaries/runtime-validation.md` | Stack-level *mechanics* of boundary validation; the new page carries the language-agnostic *decision* (what counts as a complete answer), matching AGENTS.md's common-owns-principle / stack-owns-mechanics split. No duplication |
+| `wiki/infrastructure/config/environment-config.md` | **Real overlap with insight 3** — its rule 5 ("Required keys get NO default") is the sibling case. Merged rather than duplicated: added one edge-case row (a required value naming a repo-external resource must be *resolved against the owner's catalog*, not presence-checked) pointing at the new page, plus a `related:` link; `last_verified` bumped to 2026-07-31 |
+| `wiki/qa/process/release-gates.md` | Considered as the home for insight 3 (it is a merge-time check). Not merged: release-gates is deliberately a **fixed per-release checklist**, while this is a per-PR check on one kind of diff. Linked from the new page instead, so the gate stays reachable without being diluted |
+| `grep -rn 'llm\|openai\|finish_reason\|model alias' wiki/` | Only incidental hits (`security/secrets-in-code` on pasting secrets into an LLM). No prior LLM-integration coverage anywhere |
+| `log.md` | No prior `contradiction` or overlapping ingest for these cases. The 2026-07-11 gap entry proposing `common/ingestion/external-source-sync.md` is a different case (idempotent re-ingestion) and is left untouched |
+
+**Conflicts found: none.** Nothing in the wiki contradicts these three directives.
+
+Self-lint before commit: all 3 new pages are 66–71 lines (limit 120); zero banned vague
+qualifiers (`usually|generally|consider|might want|as appropriate|typically|probably|ideally`);
+all 11 `related:`/inline page ids resolve to exactly one existing page each.
 
 ## Routing decision
 
-- Target: **`databases/query-optimization/streaming-large-result-sets.md`** (new page).
-- Category `query-optimization` fits (memory-bounding how a query's result is pulled into the app is query-execution optimization); no new category needed.
-- Registered in `databases/index.md` (query-optimization section) and appended to `log.md`.
+| Insight | Target | Rationale |
+|---------|--------|-----------|
+| 1 — non-interactive CLI invocation | `platforms` / `processes` / **new page** `non-interactive-cli-invocation.md` (`platforms-processes-non-interactive-cli-invocation`) | Existing category, no new one needed: `processes` already owns "how a process is started and kept alive outside an interactive session", and this is the sibling case (how it is *invoked*). Domain choice — the artifact changed is a shell/CI invocation, and the OS-level fact (a terminal fd 0 blocks a read) is platforms' subject; debugging is linked, not duplicated |
+| 2 — LLM response completeness | `backend` / **NEW category** `common/integrations` / `llm-response-completeness.md` | New category justified: the existing eight common categories cover the API you *publish* (api-design), transport to a dependency (reliability), and internal concerns (caching, jobs, errors, auth, orm, concurrency). None covers *depending on an externally owned service's response contract and its catalog of names*. Placed in `common/` (not `node/`/`python/`) because the decision is language-agnostic; `applies_to: [openai-compatible]` scopes it |
+| 3 — externally-owned defaults | `backend` / `common/integrations` / `externally-owned-defaults.md` | Same new category — it is the second face of one concern (insight 2 = the response you get back; insight 3 = the name you asked for). Kept as its own page under AGENTS.md's one-case-per-page rule rather than folded into insight 2, because its trigger (reviewing/merging a diff, plus startup validation) is distinct and it generalizes past LLMs to any repo-external name (bucket, queue, index). Cross-linked to `infrastructure-config-environment-config` and `qa-process-release-gates` |
+
+Plumbing updated: `wiki/backend/index.md` (new `### integrations` section, 2 rows),
+`wiki/platforms/index.md` (1 row under `processes`), `INDEX.md` (both domains' "route
+here when" lines extended so the new cases are reachable from the root map), `log.md`
+(2 `ingest` + 1 `revise` entries), and `background-services` / `environment-config`
+`related:` back-links.
+
+Queue rows for these 3 candidates are retired to `~/.dev-loop/queue/.processed.jsonl`
+once this PR is opened.
diff --git a/INDEX.md b/INDEX.md
index 05789bd..f576bbe 100644
--- a/INDEX.md
+++ b/INDEX.md
@@ -10,14 +10,14 @@ follow the cross-pointers in their index or take the next matching seeded domain
 | Domain | Status | Route here when |
 |--------|--------|-----------------|
 | [databases](wiki/databases/index.md) | **seeded** | Designing schemas/tables/keys, choosing or evaluating indexes, writing or optimizing queries, choosing transaction/isolation behavior |
-| [backend](wiki/backend/index.md) | **seeded** | Server-side application code — language-agnostic (`common/`: API contracts, idempotency, JWT, timeouts/retries, caching, jobs, transactions in app code, shared state/pools, errors) plus stack subtrees: `java/` (JPA, Spring proxies, JVM threads/memory), `node/` (event loop, promises, runtime validation, shutdown), `python/` (GIL/asyncio, pydantic, WSGI/ASGI workers, language traps) |
+| [backend](wiki/backend/index.md) | **seeded** | Server-side application code — language-agnostic (`common/`: API contracts, idempotency, JWT, timeouts/retries, caching, jobs, transactions in app code, shared state/pools, errors, consuming external/LLM APIs and defaults that name repo-external resources) plus stack subtrees: `java/` (JPA, Spring proxies, JVM threads/memory), `node/` (event loop, promises, runtime validation, shutdown), `python/` (GIL/asyncio, pydantic, WSGI/ASGI workers, language traps) |
 | [frontend](wiki/frontend/index.md) | **seeded** | Web UI code: state placement, rendering performance, in-UI data fetching (races, infinite scroll), auth token handling, forms, XSS-safe output, accessibility |
 | [infrastructure](wiki/infrastructure/index.md) | **seeded** | CI/CD pipelines, secrets in build/deploy, container image builds, rollout/rollback strategy, observability (logs/metrics/alerting) |
 | [testing](wiki/testing/index.md) | **seeded** | Writing or structuring automated tests: level choice, cases/assertions, test data, mock decisions, flaky tests (release-process quality → qa) |
 | [qa](wiki/qa/index.md) | **seeded** | Release-quality process: release gates, regression scoping, bug reports, severity/priority triage, exploratory testing (writing automated test code → testing) |
 | [debugging](wiki/debugging/index.md) | **seeded** | Diagnosing a failure — finding what is wrong and why: reproducing, bisection, hypothesis testing, traces/logs, intermittent failures (fixing the diagnosed fault → its owning domain) |
 | [security](wiki/security/index.md) | **seeded** | Trust-boundary decisions: input validation, session-vs-token auth choice, per-resource authorization (IDOR), secrets hygiene, dependency trust, PII handling (XSS rendering → frontend; CI secrets → infrastructure; JWT implementation → backend/frontend auth) |
-| [platforms](wiki/platforms/index.md) | **seeded** | OS-level differences breaking code across macOS/Linux/Windows: shell portability, BSD-vs-GNU CLI, filesystem case/line endings, background services/cron, toolchain version pinning |
+| [platforms](wiki/platforms/index.md) | **seeded** | OS-level differences breaking code across macOS/Linux/Windows: shell portability, BSD-vs-GNU CLI, filesystem case/line endings, background services/cron, invoking prompt-capable CLIs non-interactively, toolchain version pinning |
 | [mobile](wiki/mobile/index.md) | **seeded** | App-side iOS/Android/cross-platform: process death/state survival, offline-first sync, mobile-network calls, store rollout/hotfix strategy, startup time |
 
 All ten domains are seeded. New categories grow via `skills/wiki-ingest/SKILL.md`.
diff --git a/log.md b/log.md
index 587a5dc..13be087 100644
--- a/log.md
+++ b/log.md
@@ -34,3 +34,6 @@ Append-only. Format: `## [YYYY-MM-DD] <ingest|revise|lint|gap|contradiction|drif
 ## [2026-07-12] revise | security/secrets-in-code +1 edge case: third-party HTTP client (httpx/requests) logs the full request URL — including a query-param API key — at INFO, so root/DEBUG logging leaks it; keep the client logger above INFO. Found when a standalone sync process set logging.basicConfig(INFO) and httpx wrote the data.go.kr serviceKey to the log file. last_verified bumped to 2026-07-12.
 ## [2026-07-13] ingest | databases +1 (query-optimization): streaming-large-result-sets — memory-bounded export of a huge single-query result. Client-side cursor pulls the whole set to libpq on execute (fetchmany caps only the Python-list explosion); only a server-side/named cursor truly streams but needs a transaction, so it fails under autocommit or a proxy that blocks BEGIN → fall back to client-side fetchmany + disk spool + openpyxl write_only (measured 300k rows 838MB→38MB). Derived from RNR-3440 (potential-listing weekly extract memory peak); QueryPie BEGIN-block generalized to "read-only access proxy", field-tested. Sources: psycopg2 usage/cursor docs (named-cursor WITHOUT HOLD + autocommit exception), openpyxl optimized-modes (write-only near-constant memory, lxml=speed-not-memory).
 ## [2026-07-23] ingest | databases +2: schema-design/online-schema-changes (ACCESS EXCLUSIVE lock avoidance — non-volatile default fast path, ADD CONSTRAINT NOT VALID + VALIDATE at SHARE UPDATE EXCLUSIVE, CHECK-NOT-NULL trick, CREATE INDEX CONCURRENTLY, expand-and-contract to decouple DB migration from app deploy, lock_timeout for lock-queue pile-up) + operations/autovacuum-and-wraparound (NEW category operations: per-table scale_factor/cost_limit tuning for hot tables, age(datfrozenxid)/relfrozenxid + n_dead_tup monitoring, wraparound read-only cliff and superuser VACUUM recovery, VACUUM FULL vs pg_repack). Derived from the Hatchet "Postgres survival guide"; both cross-checked against PostgreSQL official docs (sql-altertable, routine-vacuuming).
+## [2026-07-31] ingest | platforms +1 (processes): non-interactive-cli-invocation — a CLI's own `-p/--print/--yes` flag does not stop it reading fd 0, so a scripted/harness call blocks on a terminal stdin and no request ever reaches the remote service; detach stdin (`</dev/null`, GNU nohup's unreadable-fd redirect), close documented prompt channels (`ssh -n`, `GIT_TERMINAL_PROMPT=0`), bound with `timeout`, then split client-vs-server by whether the request appears in the server's access log. Derived from a pi 0.79.1 foreground hang (2×, 0 bytes) with zero gateway-log arrivals that resolved instantly under nohup. Sources: GNU coreutils nohup, OpenBSD ssh(1) -n, git GIT_TERMINAL_PROMPT.
+## [2026-07-31] ingest | backend +2 (NEW category common/integrations): llm-response-completeness (HTTP 200 is transport-only; fail on `finish_reason == "length"` and blank `content`; blank content + populated `reasoning_content`/`reasoning` means a reasoning model spent the output budget — raise the cap or route to a non-reasoning alias, never publish the reasoning text; log the returned `model` beside the requested alias) + externally-owned-defaults (a default naming a resource outside the repo is re-queried against the owner's catalog at review/merge time with a dated note, and resolved at startup so a retired name fails the deploy instead of the user). Derived from a LiteLLM gateway measurement (finish=length, content 0 chars / reasoning_content 8,173 chars at 9,317 input tokens) and a PR whose default alias had disappeared from the catalog after its own end-to-end check. Sources: OpenAI reasoning guide + chat-object finish_reason + deprecations + models/list, vLLM reasoning outputs, LiteLLM reasoning_content + proxy model discovery.
+## [2026-07-31] revise | infrastructure/config/environment-config +1 edge case: a required value naming a repo-external resource must be resolved against the owner's catalog at startup, not just presence-checked; cross-linked to backend-common-integrations-externally-owned-defaults. last_verified bumped to 2026-07-31.
diff --git a/wiki/backend/common/integrations/externally-owned-defaults.md b/wiki/backend/common/integrations/externally-owned-defaults.md
new file mode 100644
index 0000000..b036b2e
--- /dev/null
+++ b/wiki/backend/common/integrations/externally-owned-defaults.md
@@ -0,0 +1,70 @@
+---
+id: backend-common-integrations-externally-owned-defaults
+domain: backend
+category: integrations
+applies_to: [general]
+confidence: verified
+sources:
+  - https://developers.openai.com/api/docs/deprecations
+  - https://developers.openai.com/api/docs/api-reference/models/list
+  - https://docs.litellm.ai/docs/proxy/model_discovery
+last_verified: 2026-07-31
+related: [backend-common-integrations-llm-response-completeness, infrastructure-config-environment-config, qa-process-release-gates]
+---
+
+# Defaults That Name a Resource Owned Outside the Repository
+
+## When this applies
+
+A code or config default names something whose existence is controlled outside your
+repository — a gateway model alias, a serving endpoint name, a bucket, a queue, a
+search index. Also when reviewing or merging a PR that claims end-to-end
+verification of such a default, and when a feature that "was tested" fails on its
+default path in an environment nobody changed.
+
+## Do this
+
+1. **Re-query the owner's catalog at review/merge time** and confirm the default
+   still resolves: `GET /v1/models` for an OpenAI-compatible provider or gateway,
+   the equivalent list/describe call otherwise. A measurement inside the PR body
+   dates from when it was written; nothing in code review re-checks a name the repo
+   does not own.
+2. **Paste the dated check into the PR** (endpoint queried, result, date), so the
+   next reviewer reads the age of the evidence instead of trusting its existence.
+3. **Resolve the configured name at startup and crash when it is absent**, with the
+   resolved name in the message — a retired alias then fails the deploy instead of
+   the user's request. Startup-validation mechanics:
+   [infrastructure-config-environment-config].
+4. **Make the runtime failure name the resource**: log the requested name and the
+   provider's error verbatim, so "400 from the gateway" reads as "alias X no longer
+   exists" without a debugging session.
+
+| Case | Do |
+|------|----|
+| The named resource is required for the feature | Validate at startup, crash on failure, and gate the deploy on that check |
+| Several names are tried in order (primary + fallbacks) | Resolve all of them at startup, log which ones exist, and alert when the primary is unresolvable while a fallback is serving |
+| The provider has announced a shutdown date | Record the date next to the default (comment + tracking issue) — preview-tier names can retire on as little as two weeks' notice |
+| The default only exists in one environment's catalog | Make it a required per-environment config value with no default ([infrastructure-config-environment-config]) |
+
+## Edge cases
+
+| Case | Then |
+|------|------|
+| CI has no credentials for the catalog | Run the check by hand at review and paste the output with its date; add the startup resolution check so runtime still catches drift |
+| Catalog lists the name but calls still fail | The name exists and your route/key lacks access — repeat the check with the exact credentials the service uses |
+| Gateway resolves wildcards or aliases to upstream models | Query the gateway's own list endpoint (LiteLLM `/v1/models`, `check_provider_endpoint` for wildcards), not the upstream provider's — the gateway's mapping is the one your code hits |
+| Default is only reached on a rarely used path | The startup check still runs — that is the point; a path exercised once a month otherwise reveals the dead name to a user first |
+
+## Instead of
+
+| If you are about to | Do this instead | Why |
+|---------------------|-----------------|-----|
+| Trust a PR's "verified end-to-end" for an externally owned name | Re-query the catalog at merge time and record the dated result | The measurement was true when written; the catalog changes outside the repo while code, tests, and build all keep passing |
+| Hardcode a convenient default so local dev "just works" | Require the value and validate it at startup ([infrastructure-config-environment-config]) | A default that suits the author's machine becomes the silent production value once the real name is retired |
+| Let the first user request discover that the name is gone | Resolve the name at startup / in the health check | The same failure costs a failed deploy instead of a broken feature plus an incident |
+
+## Sources
+
+- https://developers.openai.com/api/docs/deprecations — deprecation announces a shutdown date, after which the model "will no longer be accessible"; notice periods 6 months (GA), 3 months (specialized), as little as 2 weeks (preview)
+- https://developers.openai.com/api/docs/api-reference/models/list — `GET /v1/models` "Lists the currently available models, and provides basic information about each one such as the owner and availability"
+- https://docs.litellm.ai/docs/proxy/model_discovery — the proxy's `/v1/models` returns the models actually available behind that gateway; `check_provider_endpoint: true` resolves wildcard entries
diff --git a/wiki/backend/common/integrations/llm-response-completeness.md b/wiki/backend/common/integrations/llm-response-completeness.md
new file mode 100644
index 0000000..0a10b30
--- /dev/null
+++ b/wiki/backend/common/integrations/llm-response-completeness.md
@@ -0,0 +1,66 @@
+---
+id: backend-common-integrations-llm-response-completeness
+domain: backend
+category: integrations
+applies_to: [openai-compatible]
+confidence: verified
+sources:
+  - https://developers.openai.com/api/docs/guides/reasoning
+  - https://developers.openai.com/api/docs/api-reference/chat/object
+  - https://docs.vllm.ai/en/latest/features/reasoning_outputs.html
+  - https://docs.litellm.ai/docs/reasoning_content
+last_verified: 2026-07-31
+related: [backend-common-reliability-timeouts-and-retries, backend-common-integrations-externally-owned-defaults, backend-common-errors-exception-handling]
+---
+
+# Accepting an LLM Chat-Completion Response as a Finished Artifact
+
+## When this applies
+
+Server code posts a prompt to an OpenAI-compatible `/chat/completions` (a provider
+directly, or a gateway such as LiteLLM / vLLM) and uses the returned text as an
+artifact — persisted, posted to a channel, attached to a ticket, fed downstream.
+Which model answers is configurable, so a reasoning model can be routed in without
+a code change. Also when empty or mid-sentence output is reaching users while the
+call logs as a success.
+
+## Do this
+
+Treat HTTP 200 as "transport succeeded" only, and gate the artifact on the body:
+
+| Case | Do |
+|------|----|
+| `finish_reason == "stop"` and `content` non-blank | Accept the text |
+| `finish_reason == "length"` | Fail the call — the text stopped at your token cap mid-generation. Retry with a higher cap or a smaller input; never persist or post the partial text |
+| `content` blank (empty or whitespace) | Fail the call with a diagnostic carrying `finish_reason`, `usage` (prompt/completion/reasoning tokens), the requested alias, and the returned `model` |
+| `content` blank AND a reasoning field non-blank | Report "the reasoning model spent the output budget": raise the token cap (OpenAI advises reserving ≥25,000 tokens for reasoning + output while calibrating) or route to a non-reasoning alias. Never substitute the reasoning text for the answer — it is scratch work, not the deliverable |
+| `finish_reason == "tool_calls"` | Blank `content` is correct here — consume the tool call; do not run the blank-content failure path |
+
+Then log the `model` value the response carries next to the alias you requested, so
+a gateway reroute (alias → different model) is visible in the record afterwards.
+
+## Edge cases
+
+| Case | Then |
+|------|------|
+| Which reasoning field to read | LiteLLM returns `reasoning_content`; vLLM returns `reasoning` (renamed from `reasoning_content`). Check both keys before concluding no reasoning was produced |
+| Streaming responses | Accumulate, then apply the same gate using the final chunk's `finish_reason` — validate before the text leaves your process |
+| Caller uses the Responses API, not chat completions | The equivalent signal is `status == "incomplete"` with reason `max_output_tokens`; treat it exactly as `finish_reason == "length"` |
+| Retrying a `length` failure | This is a fresh request, not a retry of the same bytes — raise the cap or shrink the input first, otherwise it truncates identically ([backend-common-reliability-timeouts-and-retries] owns retry policy) |
+| Reasoning tokens are billed but invisible | Cost is incurred even when `content` is blank, so alert on the blank-content failure rather than silently retrying in a loop |
+
+## Instead of
+
+| If you are about to | Do this instead | Why |
+|---------------------|-----------------|-----|
+| Read `choices[0].message.content` and write it straight to the artifact | Gate on `finish_reason` and non-blank content first | A 200 with `finish_reason=length` is a truncated artifact wearing a success status; downstream systems cannot tell |
+| Treat an empty string as "the model had nothing to add" | Fail with `finish_reason` + token usage in the message | Blank content is budget exhaustion or a reasoning-only response — both are actionable configuration facts, not model opinions |
+| Raise the token cap whenever output is blank | Check the reasoning field first, then choose raise-cap vs non-reasoning route | If reasoning filled the budget, a bigger cap buys more scratch work; the alias choice is what actually decides |
+| Fall back to the reasoning text when `content` is blank | Fail the call and report the reasoning-token count | Reasoning text is unformatted intermediate state — publishing it leaks scratch work as a deliverable |
+
+## Sources
+
+- https://developers.openai.com/api/docs/guides/reasoning — reasoning tokens occupy the output budget and are billed as output tokens; a response can be cut off "before any visible output tokens are produced"; detect via `status == incomplete` / `max_output_tokens`; reserve ≥25,000 tokens while calibrating
+- https://developers.openai.com/api/docs/api-reference/chat/object — `finish_reason` values: `stop`, `length` ("the maximum number of tokens specified in the request was reached"), `tool_calls`, `content_filter`
+- https://docs.vllm.ai/en/latest/features/reasoning_outputs.html — OpenAI-compatible servers put reasoning in `reasoning` (formerly `reasoning_content`) and the final answer in `content`
+- https://docs.litellm.ai/docs/reasoning_content — LiteLLM normalizes provider thinking into `message.reasoning_content` (plus `thinking_blocks` for Anthropic)
diff --git a/wiki/backend/index.md b/wiki/backend/index.md
index e991b04..55604c8 100644
--- a/wiki/backend/index.md
+++ b/wiki/backend/index.md
@@ -32,6 +32,13 @@ Match your situation to a "load when" line; load only matching pages.
 |------|-----------|
 | [timeouts-and-retries](common/reliability/timeouts-and-retries.md) | Your service calls another service/external API/DB over the network — setting timeouts and deadlines, deciding what to retry per failure type, backoff/jitter, capping concurrency against a slow dependency; debugging pool exhaustion or retry storms |
 
+### integrations
+
+| Page | Load when |
+|------|-----------|
+| [llm-response-completeness](common/integrations/llm-response-completeness.md) | Your code consumes an OpenAI-compatible `/chat/completions` response (provider or gateway) and uses the text as an artifact — deciding what counts as a complete answer, handling `finish_reason == "length"`, blank `content`, or `reasoning_content`/`reasoning` from a reasoning model; blank or mid-sentence output reaching users while the call logs HTTP 200 |
+| [externally-owned-defaults](common/integrations/externally-owned-defaults.md) | A code/config default names a resource the repo does not own (model alias, endpoint, bucket, queue, index) — reviewing or merging a PR that claims that default works, adding a startup check that the name still resolves, or diagnosing a default path that broke with no code change |
+
 ### caching
 
 | Page | Load when |
diff --git a/wiki/infrastructure/config/environment-config.md b/wiki/infrastructure/config/environment-config.md
index 8964889..f305b16 100644
--- a/wiki/infrastructure/config/environment-config.md
+++ b/wiki/infrastructure/config/environment-config.md
@@ -8,8 +8,8 @@ sources:
   - https://12factor.net/config
   - https://12factor.net/build-release-run
   - https://12factor.net/dev-prod-parity
-last_verified: 2026-07-10
-related: [infrastructure-deploy-rollout-and-rollback, infrastructure-ci-cd-secrets-handling, backend-node-boundaries-runtime-validation, backend-python-boundaries-runtime-validation]
+last_verified: 2026-07-31
+related: [infrastructure-deploy-rollout-and-rollback, infrastructure-ci-cd-secrets-handling, backend-node-boundaries-runtime-validation, backend-python-boundaries-runtime-validation, backend-common-integrations-externally-owned-defaults]
 ---
 
 # Configuration That Differs Per Environment
@@ -62,6 +62,7 @@ files, and env vars; reviewing how a service gets its settings.
 | A key is only meaningful in prd (e.g. a payments endpoint) | Declare it required in every environment and give dev/stg a working sandbox value — an optional-in-dev key hides a missing-prd-value crash until the prd deploy |
 | A value must change without a redeploy (kill switch, tuning knob) | Use a runtime flag ([infrastructure-deploy-rollout-and-rollback] config/flag row); flag names and allowed values still belong in the schema inventory |
 | Config service or mounted config unreachable at boot | Crash and let the orchestrator restart/retry; starting with fallback values means each instance runs config you cannot account for |
+| A required value names a resource the repo does not own (model alias, endpoint, bucket, queue) | Startup validation must resolve the name against the owner's catalog, not just check that the string is present — [backend-common-integrations-externally-owned-defaults] owns the review-time and startup checks |
 | Legacy code full of `if (env === 'prod')` branches | On each touch, replace the branch you are editing with a named config value; record the remaining branches as inventory gaps |
 
 ## Instead of
diff --git a/wiki/platforms/index.md b/wiki/platforms/index.md
index dccb400..c198a7b 100644
--- a/wiki/platforms/index.md
+++ b/wiki/platforms/index.md
@@ -40,6 +40,7 @@ Match your situation to a "load when" line; load only matching pages.
 
 | Page | Load when |
 |------|-----------|
+| [non-interactive-cli-invocation](processes/non-interactive-cli-invocation.md) | Calling a CLI that can prompt (agent CLI, ssh, git, package manager) from a script/CI/hook/agent harness, including with its own `-p`/`--print`/`--yes` flag; such a call hangs with no output and no error; deciding whether a hang is client-side or the remote service; feeding input to a tool without letting it read the terminal |
 | [background-services](processes/background-services.md) | Something must run persistently or on a schedule on a dev machine or server (daemon, watcher, cron-style job); a "started" process dies when the terminal/SSH/agent session ends; choosing nohup vs LaunchAgent vs systemd unit vs cron/timer; a job works in the terminal but fails under cron/launchd (minimal environment); wiring service logs and restart policy |
 
 ## toolchains
diff --git a/wiki/platforms/processes/background-services.md b/wiki/platforms/processes/background-services.md
index f6e9503..22b1e9a 100644
--- a/wiki/platforms/processes/background-services.md
+++ b/wiki/platforms/processes/background-services.md
@@ -11,7 +11,7 @@ sources:
   - https://man7.org/linux/man-pages/man1/nohup.1.html
   - https://man7.org/linux/man-pages/man1/loginctl.1.html
 last_verified: 2026-07-10
-related: [platforms-toolchains-version-management, platforms-shells-portable-shell-scripts]
+related: [platforms-toolchains-version-management, platforms-shells-portable-shell-scripts, platforms-processes-non-interactive-cli-invocation]
 ---
 
 # Keeping a Process Running Beyond the Terminal Session
diff --git a/wiki/platforms/processes/non-interactive-cli-invocation.md b/wiki/platforms/processes/non-interactive-cli-invocation.md
new file mode 100644
index 0000000..be7f2ff
--- /dev/null
+++ b/wiki/platforms/processes/non-interactive-cli-invocation.md
@@ -0,0 +1,71 @@
+---
+id: platforms-processes-non-interactive-cli-invocation
+domain: platforms
+category: processes
+applies_to: [macos, linux]
+confidence: verified
+sources:
+  - https://www.gnu.org/software/coreutils/manual/html_node/nohup-invocation.html
+  - https://man.openbsd.org/ssh
+  - https://git-scm.com/docs/git
+last_verified: 2026-07-31
+related: [platforms-processes-background-services, platforms-tools-bsd-vs-gnu-cli, debugging-methodology-hypothesis-testing]
+---
+
+# Invoking a Prompt-Capable CLI from a Script or Agent Harness
+
+## When this applies
+
+Calling a CLI that is able to prompt (agent CLIs, `ssh`, `git`, package managers)
+from a script, CI step, hook, or agent harness — including when you passed its own
+non-interactive flag (`-p`, `--print`, `--yes`). Also when such a call hangs with no
+output and no error and you must decide whether the client, the network, or the
+remote service is at fault.
+
+## Do this
+
+1. **Detach fd 0 at the call site**: `cmd </dev/null`, or `nohup cmd >"$LOG" 2>&1 &`
+   (GNU `nohup` redirects a terminal stdin to a deliberately unreadable descriptor,
+   so a stray read reports an error instead of blocking). A non-interactive flag
+   controls the tool's output and prompt *policy*; it does not guarantee the tool
+   never reads standard input.
+2. **Also close the tool's documented prompt channel** when it has one:
+   `ssh -n` ("prevents reading from stdin … must be used when ssh is run in the
+   background"), `GIT_TERMINAL_PROMPT=0` for git's own credential prompts.
+3. **Bound every such call with a timeout** (`timeout` / `gtimeout` —
+   [platforms-tools-bsd-vs-gnu-cli] owns the macOS gap) so a blocked read fails the
+   step instead of hanging the pipeline.
+4. **Write output to a file**, so "0 bytes produced" is recorded evidence rather
+   than a memory of a blank terminal.
+5. **Split client from server before blaming the remote side**, using the server's
+   own access/gateway log for your source IP and time window:
+
+| Observation | Conclusion |
+|-------------|------------|
+| No request logged on the server for that window | The call never left the client — a local block (fd 0, auth prompt, lock). Fix the invocation |
+| Request logged, response slow or 5xx | Server side — take it to the service's latency/error path |
+| Request logged and answered fast, client still produced nothing | The client blocked *after* the response — inspect the client's own output handling |
+
+## Edge cases
+
+| Case | Then |
+|------|------|
+| The tool genuinely needs input on stdin (a piped prompt) | Feed it from a file or heredoc (`cmd <prompt.txt`) so fd 0 is a file and never the terminal |
+| Target is not GNU coreutils (macOS/BSD `nohup`) | Redirect explicitly: `nohup cmd 0>/dev/null >"$LOG" 2>&1 &` — the stdin redirect is a GNU extension |
+| Still hangs with stdin closed and nothing in the server log | Look for a non-stdin block: a lockfile, a keychain/credential prompt that only resolves in a GUI session, or a missing binary ([platforms-environment-path-resolution]) |
+| The process must outlive the session | [platforms-processes-background-services] owns detach/supervision choice |
+| Output arrives only after the process exits | The tool buffers when fd 1 is not a TTY — read the log after exit, or use the tool's line-buffered/streaming flag |
+
+## Instead of
+
+| If you are about to | Do this instead | Why |
+|---------------------|-----------------|-----|
+| Raise the client timeout because the call "is just slow" | Re-run with `</dev/null` under `timeout`, then check the server log for arrival | A blocked read on fd 0 never emits a request; more waiting cannot produce one |
+| Conclude the model/gateway/network is broken from a client-side hang | Confirm the request appears in the server's log first | Absence of the request in the server log localizes the fault to your invocation, and saves restarting healthy remote services |
+| Rely on `--print` / `-p` / `--yes` alone to make a call non-interactive | Pass the flag AND redirect stdin from `/dev/null` | The flag is the tool's intent; the redirect is the guarantee |
+
+## Sources
+
+- https://www.gnu.org/software/coreutils/manual/html_node/nohup-invocation.html — when stdin is a terminal, nohup redirects it and makes the substitute descriptor unreadable so a mistaken read reports an error; GNU extension, portable form `0>/dev/null`
+- https://man.openbsd.org/ssh — `-n` "Redirects stdin from /dev/null (actually, prevents reading from stdin). This must be used when ssh is run in the background."
+- https://git-scm.com/docs/git — `GIT_TERMINAL_PROMPT`: when false, git will not prompt on the terminal (default is to prompt)

From f0f344ff178de266a317e1004c9becbad0527248 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=EC=B5=9C=EC=98=81=EA=B8=B0?= <dch0202@rsquare.co.kr>
Date: Fri, 31 Jul 2026 11:38:01 +0900
Subject: [PATCH 2/3] knowledge: apply adversarial cross-check findings (ssh
 BatchMode, nohup precondition, function_call, routing row)

---
 .dev-loop/INGEST_REPORT.md                    | 72 ++++++++++++++++---
 log.md                                        |  3 +-
 .../integrations/externally-owned-defaults.md | 10 +--
 .../integrations/llm-response-completeness.md | 10 +--
 wiki/backend/index.md                         |  2 +-
 .../config/environment-config.md              |  2 +-
 .../non-interactive-cli-invocation.md         | 32 +++++----
 7 files changed, 98 insertions(+), 33 deletions(-)

diff --git a/.dev-loop/INGEST_REPORT.md b/.dev-loop/INGEST_REPORT.md
index c6f4745..3dcd421 100644
--- a/.dev-loop/INGEST_REPORT.md
+++ b/.dev-loop/INGEST_REPORT.md
@@ -24,8 +24,16 @@ Sources checked (all fetched live, 2026-07-31):
   background." Primary evidence that mature CLIs ship an explicit stdin-detach flag
   *because* background invocation otherwise blocks.
 - [`git` docs, `GIT_TERMINAL_PROMPT`](https://git-scm.com/docs/git) — "If this Boolean
-  environment variable is set to false, git will not prompt on the terminal", i.e.
-  prompting on the terminal is the default. Second independent instance of the pattern.
+  environment variable is set to false, git will not prompt on the terminal (e.g., when
+  asking for HTTP authentication)." Second independent instance of a CLI shipping an
+  explicit opt-out for terminal prompting. (The doc does **not** state the default; an
+  earlier draft asserted "default is to prompt" as sourced — removed by cross-check.)
+- [OpenBSD `ssh_config`](https://man.openbsd.org/ssh_config) — `BatchMode=yes`: "user
+  interaction such as password prompts and host key confirmation requests will be
+  disabled"; `StrictHostKeyChecking=accept-new` adds new host keys without permitting
+  changed ones. Added by cross-check: `ssh -n` detaches stdin but the same man page notes
+  it "does not work if ssh needs to ask for a password or passphrase", so the prompt
+  channel needs its own directive.
 
 Reproduction from the session: `pi 0.79.1 --tools` in the foreground hung twice
 (300s, 150s; 0 bytes written), while the LiteLLM gateway log showed **zero requests from
@@ -50,8 +58,10 @@ Sources checked (all fetched live, 2026-07-31):
   ≥25,000 tokens while calibrating. This is exactly the failure mode the candidate claimed.
 - [OpenAI chat-completion object](https://developers.openai.com/api/docs/api-reference/chat/object)
   — `finish_reason: "length"` = "the maximum number of tokens specified in the request
-  was reached"; `tool_calls` and `content_filter` are the other non-`stop` values (the
-  `tool_calls` case is why the page carves out legitimately-blank content).
+  was reached"; the full set is five values — `stop`, `length`, `tool_calls`,
+  `content_filter`, and the deprecated `function_call`. Both tool-calling values return
+  blank `content` by design, which is why the page carves them out of the
+  blank-content failure path (the `function_call` half was added by cross-check).
 - [vLLM reasoning outputs](https://docs.vllm.ai/en/latest/features/reasoning_outputs.html)
   — reasoning goes in `reasoning` (renamed from `reasoning_content`), final answer in
   `content`.
@@ -115,7 +125,8 @@ Pages read in full before writing anything:
 
 **Conflicts found: none.** Nothing in the wiki contradicts these three directives.
 
-Self-lint before commit: all 3 new pages are 66–71 lines (limit 120); zero banned vague
+Self-lint before commit (re-run after the cross-check fixes): all 3 new pages are 66–77
+lines (limit 120); zero banned vague
 qualifiers (`usually|generally|consider|might want|as appropriate|typically|probably|ideally`);
 all 11 `related:`/inline page ids resolve to exactly one existing page each.
 
@@ -127,11 +138,56 @@ all 11 `related:`/inline page ids resolve to exactly one existing page each.
 | 2 — LLM response completeness | `backend` / **NEW category** `common/integrations` / `llm-response-completeness.md` | New category justified: the existing eight common categories cover the API you *publish* (api-design), transport to a dependency (reliability), and internal concerns (caching, jobs, errors, auth, orm, concurrency). None covers *depending on an externally owned service's response contract and its catalog of names*. Placed in `common/` (not `node/`/`python/`) because the decision is language-agnostic; `applies_to: [openai-compatible]` scopes it |
 | 3 — externally-owned defaults | `backend` / `common/integrations` / `externally-owned-defaults.md` | Same new category — it is the second face of one concern (insight 2 = the response you get back; insight 3 = the name you asked for). Kept as its own page under AGENTS.md's one-case-per-page rule rather than folded into insight 2, because its trigger (reviewing/merging a diff, plus startup validation) is distinct and it generalizes past LLMs to any repo-external name (bucket, queue, index). Cross-linked to `infrastructure-config-environment-config` and `qa-process-release-gates` |
 
-Plumbing updated: `wiki/backend/index.md` (new `### integrations` section, 2 rows),
+## Cross-check
+
+Cross-Check: independent headless `claude -p --permission-mode plan` adversarial review of
+all three pages + this report returned **VERDICT: FAIL with 12 findings**; 8 were confirmed
+against the primary sources and applied, 2 were already clean, 2 were declined with reasons.
+Applied:
+
+1. **`ssh -n` was miscategorised** as a prompt-channel control. ssh(1) itself says `-n`
+   "does not work if ssh needs to ask for a password or passphrase" — it is a stdin detach.
+   Moved to directive 1; directive 2 now uses `-o BatchMode=yes` (+
+   `StrictHostKeyChecking=accept-new`), verified against ssh_config and cited.
+2. **GNU nohup's stdin redirect is conditional** ("If standard input is a terminal…"). In
+   this page's own CI/hook triggers fd 0 is often a pipe, so nohup redirects nothing. The
+   page now states that `</dev/null` is the unconditional guarantee.
+3. **"No request in the server log ⇒ never left the client" was too strong** — DNS, TLS, and
+   proxy failures also produce no access-log entry. Row and `Instead of` row widened to
+   "pre-log rejection", with `curl -v` as the discriminator.
+4. **Deprecated `function_call` also returns blank `content`** by design and would have hit
+   the fail path. Added to the carve-out.
+5. **The `finish_reason` enumeration was incomplete** (4 of 5) in both the page's Sources
+   bullet and this report. Corrected in both.
+6. **`GIT_TERMINAL_PROMPT` "default is to prompt" was inference presented as sourced** — the
+   doc states only the false case. Removed.
+7. **`wiki/backend/index.md`'s concern-first subtree row was not updated** — routing protocol
+   step 2 reads that row before the category tables, so an agent would have landed on
+   `reliability` instead of the new `integrations` category. Row extended (the real routing
+   defect in this batch).
+8. **`environment-config`'s `last_verified` bump was unearned** — a one-row edge-case addition
+   does not re-verify its three 12factor sources. Reverted to 2026-07-10 so lint's staleness
+   clock is not reset.
+
+Also applied for format: four anti-patterns phrased as negations inside `Do this`/`Edge cases`
+tables were restated positively (AGENTS.md keeps negations in `Instead of`), and
+`externally-owned-defaults`' pure-pointer row and startup directive were given app-side
+substance.
+
+Declined, with reasons: (a) **splitting `externally-owned-defaults` by trigger** — the
+corpus norm (`environment-config`, `release-gates`) is several related triggers plus a
+numbered list *and* a case table on one page, and the three triggers here are one case seen
+at review time, boot time, and failure time; (b) **relocating it to `infrastructure/config`**
+— its primary trigger is reviewing application-code defaults and its resolution rule lives
+in the client wrapper, so `backend` owns it; the config-shape half is delegated by link and
+`environment-config` carries the reciprocal row.
+
+Plumbing updated: `wiki/backend/index.md` (new `### integrations` section, 2 rows, plus the
+concern-first subtree row),
 `wiki/platforms/index.md` (1 row under `processes`), `INDEX.md` (both domains' "route
 here when" lines extended so the new cases are reachable from the root map), `log.md`
-(2 `ingest` + 1 `revise` entries), and `background-services` / `environment-config`
-`related:` back-links.
+(2 `ingest` + 2 `revise` entries, the second recording the cross-check), and
+`background-services` / `environment-config` `related:` back-links.
 
 Queue rows for these 3 candidates are retired to `~/.dev-loop/queue/.processed.jsonl`
 once this PR is opened.
diff --git a/log.md b/log.md
index 13be087..f72be48 100644
--- a/log.md
+++ b/log.md
@@ -36,4 +36,5 @@ Append-only. Format: `## [YYYY-MM-DD] <ingest|revise|lint|gap|contradiction|drif
 ## [2026-07-23] ingest | databases +2: schema-design/online-schema-changes (ACCESS EXCLUSIVE lock avoidance — non-volatile default fast path, ADD CONSTRAINT NOT VALID + VALIDATE at SHARE UPDATE EXCLUSIVE, CHECK-NOT-NULL trick, CREATE INDEX CONCURRENTLY, expand-and-contract to decouple DB migration from app deploy, lock_timeout for lock-queue pile-up) + operations/autovacuum-and-wraparound (NEW category operations: per-table scale_factor/cost_limit tuning for hot tables, age(datfrozenxid)/relfrozenxid + n_dead_tup monitoring, wraparound read-only cliff and superuser VACUUM recovery, VACUUM FULL vs pg_repack). Derived from the Hatchet "Postgres survival guide"; both cross-checked against PostgreSQL official docs (sql-altertable, routine-vacuuming).
 ## [2026-07-31] ingest | platforms +1 (processes): non-interactive-cli-invocation — a CLI's own `-p/--print/--yes` flag does not stop it reading fd 0, so a scripted/harness call blocks on a terminal stdin and no request ever reaches the remote service; detach stdin (`</dev/null`, GNU nohup's unreadable-fd redirect), close documented prompt channels (`ssh -n`, `GIT_TERMINAL_PROMPT=0`), bound with `timeout`, then split client-vs-server by whether the request appears in the server's access log. Derived from a pi 0.79.1 foreground hang (2×, 0 bytes) with zero gateway-log arrivals that resolved instantly under nohup. Sources: GNU coreutils nohup, OpenBSD ssh(1) -n, git GIT_TERMINAL_PROMPT.
 ## [2026-07-31] ingest | backend +2 (NEW category common/integrations): llm-response-completeness (HTTP 200 is transport-only; fail on `finish_reason == "length"` and blank `content`; blank content + populated `reasoning_content`/`reasoning` means a reasoning model spent the output budget — raise the cap or route to a non-reasoning alias, never publish the reasoning text; log the returned `model` beside the requested alias) + externally-owned-defaults (a default naming a resource outside the repo is re-queried against the owner's catalog at review/merge time with a dated note, and resolved at startup so a retired name fails the deploy instead of the user). Derived from a LiteLLM gateway measurement (finish=length, content 0 chars / reasoning_content 8,173 chars at 9,317 input tokens) and a PR whose default alias had disappeared from the catalog after its own end-to-end check. Sources: OpenAI reasoning guide + chat-object finish_reason + deprecations + models/list, vLLM reasoning outputs, LiteLLM reasoning_content + proxy model discovery.
-## [2026-07-31] revise | infrastructure/config/environment-config +1 edge case: a required value naming a repo-external resource must be resolved against the owner's catalog at startup, not just presence-checked; cross-linked to backend-common-integrations-externally-owned-defaults. last_verified bumped to 2026-07-31.
+## [2026-07-31] revise | infrastructure/config/environment-config +1 edge case: a required value naming a repo-external resource must be resolved against the owner's catalog at startup, not just presence-checked; cross-linked to backend-common-integrations-externally-owned-defaults. `last_verified` left at 2026-07-10 — the page's own 12factor sources were not re-verified for this one-row addition (adversarial cross-check caught the stale-clock reset).
+## [2026-07-31] revise | Adversarial cross-check (independent headless claude, VERDICT FAIL → 8 defects applied) on the three pages above: `ssh -n` reclassified from prompt-suppression to stdin-detach with `BatchMode=yes` (+`StrictHostKeyChecking=accept-new`) added as the prompt-channel directive and ssh_config cited; GNU nohup's stdin redirect qualified as terminal-only so the explicit `</dev/null` carries the guarantee; "no request in the server log" widened from "never left the client" to include pre-log rejection (DNS/TLS/proxy) with a `curl -v` discriminator; deprecated `function_call` added to the blank-content carve-out and the five-value `finish_reason` enumeration corrected; the GIT_TERMINAL_PROMPT "default is to prompt" parenthetical dropped as unsourced inference; backend/index.md concern-first subtree row extended (an agent routing by that line, read before the category tables, would otherwise land on `reliability`); externally-owned-defaults' pure-pointer row and startup directive given app-side substance. Declined: splitting externally-owned-defaults per trigger, and relocating it to infrastructure/config — its primary trigger is reviewing application-code defaults and the corpus already mixes numbered lists with case tables.
diff --git a/wiki/backend/common/integrations/externally-owned-defaults.md b/wiki/backend/common/integrations/externally-owned-defaults.md
index b036b2e..b262613 100644
--- a/wiki/backend/common/integrations/externally-owned-defaults.md
+++ b/wiki/backend/common/integrations/externally-owned-defaults.md
@@ -31,9 +31,11 @@ default path in an environment nobody changed.
    does not own.
 2. **Paste the dated check into the PR** (endpoint queried, result, date), so the
    next reviewer reads the age of the evidence instead of trusting its existence.
-3. **Resolve the configured name at startup and crash when it is absent**, with the
-   resolved name in the message — a retired alias then fails the deploy instead of
-   the user's request. Startup-validation mechanics:
+3. **Resolve the configured name at startup, in the client wrapper that owns the
+   call, and crash when it is absent** — the process that will use the name is the
+   one that must prove it exists, so a retired alias fails the deploy instead of the
+   user's request. Put the resolution in the same module that builds the request, so
+   a new call site cannot bypass it. Config-shape and startup-validation mechanics:
    [infrastructure-config-environment-config].
 4. **Make the runtime failure name the resource**: log the requested name and the
    provider's error verbatim, so "400 from the gateway" reads as "alias X no longer
@@ -44,7 +46,7 @@ default path in an environment nobody changed.
 | The named resource is required for the feature | Validate at startup, crash on failure, and gate the deploy on that check |
 | Several names are tried in order (primary + fallbacks) | Resolve all of them at startup, log which ones exist, and alert when the primary is unresolvable while a fallback is serving |
 | The provider has announced a shutdown date | Record the date next to the default (comment + tracking issue) — preview-tier names can retire on as little as two weeks' notice |
-| The default only exists in one environment's catalog | Make it a required per-environment config value with no default ([infrastructure-config-environment-config]) |
+| The name exists in one environment's catalog but not another's | Make it a required per-environment value with no default, and run the resolution check in every environment at boot — one shared default silently points staging at a name only production owns ([infrastructure-config-environment-config] owns the config shape) |
 
 ## Edge cases
 
diff --git a/wiki/backend/common/integrations/llm-response-completeness.md b/wiki/backend/common/integrations/llm-response-completeness.md
index 0a10b30..68ae0ba 100644
--- a/wiki/backend/common/integrations/llm-response-completeness.md
+++ b/wiki/backend/common/integrations/llm-response-completeness.md
@@ -31,10 +31,10 @@ Treat HTTP 200 as "transport succeeded" only, and gate the artifact on the body:
 | Case | Do |
 |------|----|
 | `finish_reason == "stop"` and `content` non-blank | Accept the text |
-| `finish_reason == "length"` | Fail the call — the text stopped at your token cap mid-generation. Retry with a higher cap or a smaller input; never persist or post the partial text |
+| `finish_reason == "length"` | Fail the call — the text stopped at your token cap mid-generation. Retry with a higher cap or a smaller input, and keep the partial text out of the artifact |
 | `content` blank (empty or whitespace) | Fail the call with a diagnostic carrying `finish_reason`, `usage` (prompt/completion/reasoning tokens), the requested alias, and the returned `model` |
-| `content` blank AND a reasoning field non-blank | Report "the reasoning model spent the output budget": raise the token cap (OpenAI advises reserving ≥25,000 tokens for reasoning + output while calibrating) or route to a non-reasoning alias. Never substitute the reasoning text for the answer — it is scratch work, not the deliverable |
-| `finish_reason == "tool_calls"` | Blank `content` is correct here — consume the tool call; do not run the blank-content failure path |
+| `content` blank AND a reasoning field non-blank | Report "the reasoning model spent the output budget": raise the token cap (OpenAI advises reserving ≥25,000 tokens for reasoning + output while calibrating) or route to a non-reasoning alias. Source the answer from `content` alone — the reasoning field is scratch work, not the deliverable |
+| `finish_reason == "tool_calls"`, or the deprecated `function_call` | Blank `content` is correct here — consume the tool call and skip the blank-content row above |
 
 Then log the `model` value the response carries next to the alias you requested, so
 a gateway reroute (alias → different model) is visible in the record afterwards.
@@ -47,7 +47,7 @@ a gateway reroute (alias → different model) is visible in the record afterward
 | Streaming responses | Accumulate, then apply the same gate using the final chunk's `finish_reason` — validate before the text leaves your process |
 | Caller uses the Responses API, not chat completions | The equivalent signal is `status == "incomplete"` with reason `max_output_tokens`; treat it exactly as `finish_reason == "length"` |
 | Retrying a `length` failure | This is a fresh request, not a retry of the same bytes — raise the cap or shrink the input first, otherwise it truncates identically ([backend-common-reliability-timeouts-and-retries] owns retry policy) |
-| Reasoning tokens are billed but invisible | Cost is incurred even when `content` is blank, so alert on the blank-content failure rather than silently retrying in a loop |
+| Reasoning tokens are billed but invisible | Cost is incurred even when `content` is blank — send the blank-content failure to an alert with the token counts attached, and cap automatic retries at one |
 
 ## Instead of
 
@@ -61,6 +61,6 @@ a gateway reroute (alias → different model) is visible in the record afterward
 ## Sources
 
 - https://developers.openai.com/api/docs/guides/reasoning — reasoning tokens occupy the output budget and are billed as output tokens; a response can be cut off "before any visible output tokens are produced"; detect via `status == incomplete` / `max_output_tokens`; reserve ≥25,000 tokens while calibrating
-- https://developers.openai.com/api/docs/api-reference/chat/object — `finish_reason` values: `stop`, `length` ("the maximum number of tokens specified in the request was reached"), `tool_calls`, `content_filter`
+- https://developers.openai.com/api/docs/api-reference/chat/object — five documented `finish_reason` values: `stop`, `length` ("the maximum number of tokens specified in the request was reached"), `tool_calls`, `content_filter`, and the deprecated `function_call` — the two tool-calling values are why blank `content` is carved out
 - https://docs.vllm.ai/en/latest/features/reasoning_outputs.html — OpenAI-compatible servers put reasoning in `reasoning` (formerly `reasoning_content`) and the final answer in `content`
 - https://docs.litellm.ai/docs/reasoning_content — LiteLLM normalizes provider thinking into `message.reasoning_content` (plus `thinking_blocks` for Anthropic)
diff --git a/wiki/backend/index.md b/wiki/backend/index.md
index 55604c8..8bfe517 100644
--- a/wiki/backend/index.md
+++ b/wiki/backend/index.md
@@ -5,7 +5,7 @@ three stack subtrees — route by concern first, stack second:
 
 | Subtree | Route there when |
 |---------|------------------|
-| [common](#common-language-agnostic) (below) | The concern is language-agnostic: API contracts, idempotency, JWT issuance, outbound calls, caching, jobs, transactions in app code, shared state/pools, exception structure |
+| [common](#common-language-agnostic) (below) | The concern is language-agnostic: API contracts, idempotency, JWT issuance, outbound calls, consuming an external/LLM API's response contract, defaults naming repo-external resources, caching, jobs, transactions in app code, shared state/pools, exception structure |
 | [java](java/index.md) | You are writing/reviewing JVM backend code (Java/Kotlin, Spring, JPA/Hibernate) and the concern is stack-specific: entity mapping, persistence context, proxy pitfalls, JVM threads/memory |
 | [node](node/index.md) | You are writing/reviewing Node.js/TypeScript backend code: event-loop blocking, promise error handling, runtime validation at boundaries, graceful shutdown |
 | [python](python/index.md) | You are writing/reviewing Python backend code: GIL/concurrency model, pydantic validation, WSGI/ASGI workers, language traps |
diff --git a/wiki/infrastructure/config/environment-config.md b/wiki/infrastructure/config/environment-config.md
index f305b16..3f843d4 100644
--- a/wiki/infrastructure/config/environment-config.md
+++ b/wiki/infrastructure/config/environment-config.md
@@ -8,7 +8,7 @@ sources:
   - https://12factor.net/config
   - https://12factor.net/build-release-run
   - https://12factor.net/dev-prod-parity
-last_verified: 2026-07-31
+last_verified: 2026-07-10
 related: [infrastructure-deploy-rollout-and-rollback, infrastructure-ci-cd-secrets-handling, backend-node-boundaries-runtime-validation, backend-python-boundaries-runtime-validation, backend-common-integrations-externally-owned-defaults]
 ---
 
diff --git a/wiki/platforms/processes/non-interactive-cli-invocation.md b/wiki/platforms/processes/non-interactive-cli-invocation.md
index be7f2ff..d804f61 100644
--- a/wiki/platforms/processes/non-interactive-cli-invocation.md
+++ b/wiki/platforms/processes/non-interactive-cli-invocation.md
@@ -7,6 +7,7 @@ confidence: verified
 sources:
   - https://www.gnu.org/software/coreutils/manual/html_node/nohup-invocation.html
   - https://man.openbsd.org/ssh
+  - https://man.openbsd.org/ssh_config
   - https://git-scm.com/docs/git
 last_verified: 2026-07-31
 related: [platforms-processes-background-services, platforms-tools-bsd-vs-gnu-cli, debugging-methodology-hypothesis-testing]
@@ -24,14 +25,18 @@ remote service is at fault.
 
 ## Do this
 
-1. **Detach fd 0 at the call site**: `cmd </dev/null`, or `nohup cmd >"$LOG" 2>&1 &`
-   (GNU `nohup` redirects a terminal stdin to a deliberately unreadable descriptor,
-   so a stray read reports an error instead of blocking). A non-interactive flag
-   controls the tool's output and prompt *policy*; it does not guarantee the tool
-   never reads standard input.
-2. **Also close the tool's documented prompt channel** when it has one:
-   `ssh -n` ("prevents reading from stdin … must be used when ssh is run in the
-   background"), `GIT_TERMINAL_PROMPT=0` for git's own credential prompts.
+1. **Detach fd 0 at the call site**: `cmd </dev/null` — the redirect is the
+   unconditional guarantee, and the tool's own stdin-detach flag where it has one
+   (`ssh -n` "prevents reading from stdin … must be used when ssh is run in the
+   background"). GNU `nohup` adds this **only when fd 0 is already a terminal**, so
+   in a CI step or hook where fd 0 is a pipe, write the redirect explicitly. A
+   non-interactive flag controls the tool's output and prompt *policy*; it does not
+   guarantee the tool never reads standard input.
+2. **Close the tool's prompt channel separately** — detaching stdin makes a prompt
+   fail rather than answer it: `ssh -o BatchMode=yes` ("user interaction such as
+   password prompts and host key confirmation requests will be disabled", plus
+   `-o StrictHostKeyChecking=accept-new` for a first-contact host key),
+   `GIT_TERMINAL_PROMPT=0` for git's credential prompts.
 3. **Bound every such call with a timeout** (`timeout` / `gtimeout` —
    [platforms-tools-bsd-vs-gnu-cli] owns the macOS gap) so a blocked read fails the
    step instead of hanging the pipeline.
@@ -42,7 +47,7 @@ remote service is at fault.
 
 | Observation | Conclusion |
 |-------------|------------|
-| No request logged on the server for that window | The call never left the client — a local block (fd 0, auth prompt, lock). Fix the invocation |
+| No request logged on the server for that window | The request never reached the server's logged surface: either a local block (fd 0, auth prompt, lock) or a pre-log rejection (DNS, TLS handshake, an intermediate proxy). Separate the two by retrying the same endpoint with a minimal client (`curl -v`) from the same host |
 | Request logged, response slow or 5xx | Server side — take it to the service's latency/error path |
 | Request logged and answered fast, client still produced nothing | The client blocked *after* the response — inspect the client's own output handling |
 
@@ -61,11 +66,12 @@ remote service is at fault.
 | If you are about to | Do this instead | Why |
 |---------------------|-----------------|-----|
 | Raise the client timeout because the call "is just slow" | Re-run with `</dev/null` under `timeout`, then check the server log for arrival | A blocked read on fd 0 never emits a request; more waiting cannot produce one |
-| Conclude the model/gateway/network is broken from a client-side hang | Confirm the request appears in the server's log first | Absence of the request in the server log localizes the fault to your invocation, and saves restarting healthy remote services |
+| Conclude the model/gateway/network is broken from a client-side hang | Confirm the request appears in the server's log first, then re-test the same endpoint with `curl -v` | Absence in the log narrows the fault to your side of the logged surface (invocation, DNS, TLS, proxy) and saves restarting healthy remote services |
 | Rely on `--print` / `-p` / `--yes` alone to make a call non-interactive | Pass the flag AND redirect stdin from `/dev/null` | The flag is the tool's intent; the redirect is the guarantee |
 
 ## Sources
 
-- https://www.gnu.org/software/coreutils/manual/html_node/nohup-invocation.html — when stdin is a terminal, nohup redirects it and makes the substitute descriptor unreadable so a mistaken read reports an error; GNU extension, portable form `0>/dev/null`
-- https://man.openbsd.org/ssh — `-n` "Redirects stdin from /dev/null (actually, prevents reading from stdin). This must be used when ssh is run in the background."
-- https://git-scm.com/docs/git — `GIT_TERMINAL_PROMPT`: when false, git will not prompt on the terminal (default is to prompt)
+- https://www.gnu.org/software/coreutils/manual/html_node/nohup-invocation.html — "If standard input is a terminal, redirect it … Make the substitute file descriptor unreadable, so that commands that mistakenly attempt to read from standard input can report an error"; GNU extension, portable form `0>/dev/null`. The terminal precondition is why the explicit redirect is the guarantee
+- https://man.openbsd.org/ssh — `-n` "Redirects stdin from /dev/null (actually, prevents reading from stdin). This must be used when ssh is run in the background"; the same entry notes it "does not work if ssh needs to ask for a password or passphrase" — a stdin detach, not a prompt suppressor
+- https://man.openbsd.org/ssh_config — `BatchMode=yes`: "user interaction such as password prompts and host key confirmation requests will be disabled"; `StrictHostKeyChecking=accept-new` adds new host keys without permitting changed ones
+- https://git-scm.com/docs/git — `GIT_TERMINAL_PROMPT`: "If this Boolean environment variable is set to false, git will not prompt on the terminal (e.g., when asking for HTTP authentication)"

From 940fa11ac31a9692421baa21f09d9decfabe2d12 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=EC=B5=9C=EC=98=81=EA=B8=B0?= <dch0202@rsquare.co.kr>
Date: Fri, 31 Jul 2026 11:39:09 +0900
Subject: [PATCH 3/3] knowledge: add decision log to ingest report

---
 .dev-loop/INGEST_REPORT.md | 27 +++++++++++++++++++++++++++
 1 file changed, 27 insertions(+)

diff --git a/.dev-loop/INGEST_REPORT.md b/.dev-loop/INGEST_REPORT.md
index 3dcd421..48af772 100644
--- a/.dev-loop/INGEST_REPORT.md
+++ b/.dev-loop/INGEST_REPORT.md
@@ -191,3 +191,30 @@ here when" lines extended so the new cases are reachable from the root map), `lo
 
 Queue rows for these 3 candidates are retired to `~/.dev-loop/queue/.processed.jsonl`
 once this PR is opened.
+
+## Decision Log (AI 생성)
+
+### 의도 — 무엇을 / 왜
+- 큐에 쌓인 ★ Insight 3건을 **검증 후** 위키로 승격. 원문(GNU nohup·OpenBSD ssh/ssh_config·git·OpenAI reasoning/chat-object/deprecations/models·vLLM·LiteLLM)을 직접 열어 세 directive 모두 `confidence: verified`로 확정했고, 후보에 없던 사실(게이트웨이별 `reasoning` vs `reasoning_content` 필드명 차이, 게이트웨이 자체 카탈로그를 조회해야 함)은 검증 과정에서 발견해 반영.
+- 라우팅은 **merge-before-create**: 인사이트 3은 `infrastructure/config/environment-config`의 "필수 키엔 default 없음" 규칙과 형제 케이스라 그 페이지에 edge case 1행 + 양방향 `related`로 연결하고, 본체는 리뷰시점 체크라는 별개 트리거이므로 별 페이지로 분리.
+- `backend/common/integrations`를 **새 카테고리로 신설** — 기존 8개 공통 카테고리는 "내가 발행하는 API"(api-design)와 "전송"(reliability)만 다루고, *외부 소유 서비스의 응답 스펙·이름 카탈로그에 의존하는 것*을 담을 자리가 없었다.
+- PR 본문 = INGEST_REPORT (게이트가 요구하는 3개 섹션 + 검증 근거를 리뷰어가 한 화면에서 보게).
+
+### 배제한 대안 — 무엇을 안 했나 / 왜
+- **인사이트 1을 `debugging`에 두기** — 클라이언트/서버 가르기가 진단이라 후보였지만, 바뀌는 산출물이 *호출 방식*이고 "터미널 fd 0은 읽기를 막는다"는 OS 사실이라 `platforms`. 진단 단계는 `debugging-methodology-hypothesis-testing` 링크로 처리(중복 금지).
+- **인사이트 3을 `qa/process/release-gates`에 병합** — release-gates는 의도적으로 *릴리즈마다 동일한 고정 체크리스트*인데 이건 특정 diff 유형에 걸리는 per-PR 체크라 게이트를 희석시킴. 링크만.
+- **인사이트 3을 `infrastructure/config`로 이전**(적대검증 제안) — 거절. 1차 트리거가 애플리케이션 코드 default 리뷰이고 해소 지점이 호출을 만드는 client wrapper라 `backend` 소유가 맞음. config 형태는 링크로 위임.
+- **인사이트 2·3을 한 페이지로 합치기** — AGENTS.md의 one-case-per-page 위반(트리거가 각각 응답 수신 / 리뷰·부팅 시점).
+- **`last_verified` 일괄 bump** — 12factor 출처를 재검증하지 않았으므로 `environment-config`는 2026-07-10 유지(적대검증 지적 반영).
+
+### 리뷰어가 볼 곳 — 신뢰성 판단 포인트
+- `wiki/platforms/processes/non-interactive-cli-invocation.md:26-38` — `ssh -n`(stdin 분리)과 `BatchMode=yes`(프롬프트 차단)를 분리한 부분. 1차 초안은 `-n`을 프롬프트 차단으로 잘못 분류했고 적대검증이 man page 인용으로 잡음.
+- 같은 파일 `:45`, `:64` — "서버 로그에 요청 없음"의 결론 범위. DNS/TLS/프록시 선차단을 포함하도록 약화했는데, 이 정도가 적정한지 판단 필요.
+- `wiki/backend/common/integrations/llm-response-completeness.md:31-38` — 수용/실패 판정 테이블. 특히 `tool_calls`/deprecated `function_call`에서 blank content가 정상인 carve-out.
+- `wiki/backend/index.md:8` — concern-first subtree 행. 라우팅 프로토콜 2단계가 카테고리 테이블보다 **먼저** 읽는 줄이라, 여기 누락되면 새 카테고리에 도달하지 못한다(적대검증이 잡은 유일한 실제 라우팅 결함).
+- `wiki/backend/common/integrations/externally-owned-defaults.md` 도메인 선택 — 위 "배제한 대안"의 판단이 맞는지가 이 PR에서 가장 논쟁적인 지점.
+- `log.md` 마지막 2줄 — 적대검증 결과와 거절 항목을 기록으로 남긴 부분.
+
+> [추정] 표시 항목 없음 — 모든 의도는 이 세션의 검증·적대검증 기록에 근거가 있다.
+>
+> 컨펌 예외: 이 PR은 auto-flush(headless) 실행으로 생성되어 푸시 전 대면 컨펌을 받지 못했다. 대신 ① 푸시 대상은 본인 fork(`dch0202-rsquare/dev-loop`)이고 ② 머지는 하지 않으며 ③ 리뷰·머지 판단 전체를 레포 오너에게 남긴다.