From ca0e278fec9637d1f27983fd57499c79d94e90ec Mon Sep 17 00:00:00 2001 From: lbruton Date: Thu, 5 Nov 2026 12:03:21 -0600 Subject: [PATCH 1/3] chore: document the timeline include_subagents default in the Overview --- DocVault/Overview.md | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/DocVault/Overview.md b/DocVault/Overview.md index 1fb0674..b41e1bd 100644 --- a/DocVault/Overview.md +++ b/DocVault/Overview.md @@ -94,7 +94,8 @@ Each result names the signals that ranked it (SESF-68, SESF-77): `Semantic: 0.44 Ingest scans turn text for tracker issue tokens. The deduplicated, uppercased set goes in a structured `issue_ids` field (Milvus `VARCHAR(4096)` plus an FTS5 `issue_ids` metadata column). `issue_patterns.py` decides which tokens count. The dashed preset is always on: `[A-Z][A-Z0-9]+-\d+`, boundary-anchored, with a technical-standard prefix denylist. `~/.sessionflow/issue_ids.conf`, or the path in `SESSIONFLOW_ISSUE_ID_PATTERNS`, adds the `servicenow` preset, custom `regex:` lines and `deny:` prefixes (see `issue_ids.conf.example`). - **`issue_id` filter** on `search_all_sessions` / `search_session` — optional structured exact-token pre-filter (`issue_ids like "%,SESF-25,%"`), case-insensitive, combinable conjunctively with `provider` / `project_root` / `date_from`. Unset leaves search behavior byte-identical. -- **`get_issue_timeline(issue_id, limit=50, provider, date_from, date_to)`** MCP tool + `GET /timeline` HTTP endpoint — single deduplicated, oldest-first cross-harness feed of every turn referencing an issue. Unions the structured field with an FTS keyword fallback (covering un-tagged turns), deduped by `doc_id`, sorted by `(timestamp, doc_id)`. +- **`get_issue_timeline(issue_id, limit=50, provider, date_from, date_to, include_subagents=false)`** MCP tool + `GET /timeline` HTTP endpoint — single deduplicated, oldest-first cross-harness feed of every turn referencing an issue. Unions the structured field with an FTS keyword fallback (covering un-tagged turns), deduped by `doc_id`, sorted by `(timestamp, doc_id)`. +- The timeline leaves out `agent-*` sessions (subagents and compactions) unless `include_subagents` is true. The output names the number of hidden turns. `/timeline` returns them as `hidden_subagents` and `hidden_subagents_capped`. - **Re-tagging stored rows:** `cleanup.py retag-issues` re-extracts issue IDs from every stored `document` with the active patterns and rewrites the changed rows in Milvus and FTS without re-embedding. It is a dry run by default. Run `--apply --yes` after a config change, with the server stopped. ## Specs From fa56cc69b886c3862120c11e02b8a9271b987174 Mon Sep 17 00:00:00 2001 From: lbruton Date: Thu, 5 Nov 2026 12:03:33 -0600 Subject: [PATCH 2/3] chore: replace the em-dash on the timeline bullet --- DocVault/Overview.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/DocVault/Overview.md b/DocVault/Overview.md index b41e1bd..3534c9c 100644 --- a/DocVault/Overview.md +++ b/DocVault/Overview.md @@ -94,7 +94,7 @@ Each result names the signals that ranked it (SESF-68, SESF-77): `Semantic: 0.44 Ingest scans turn text for tracker issue tokens. The deduplicated, uppercased set goes in a structured `issue_ids` field (Milvus `VARCHAR(4096)` plus an FTS5 `issue_ids` metadata column). `issue_patterns.py` decides which tokens count. The dashed preset is always on: `[A-Z][A-Z0-9]+-\d+`, boundary-anchored, with a technical-standard prefix denylist. `~/.sessionflow/issue_ids.conf`, or the path in `SESSIONFLOW_ISSUE_ID_PATTERNS`, adds the `servicenow` preset, custom `regex:` lines and `deny:` prefixes (see `issue_ids.conf.example`). - **`issue_id` filter** on `search_all_sessions` / `search_session` — optional structured exact-token pre-filter (`issue_ids like "%,SESF-25,%"`), case-insensitive, combinable conjunctively with `provider` / `project_root` / `date_from`. Unset leaves search behavior byte-identical. -- **`get_issue_timeline(issue_id, limit=50, provider, date_from, date_to, include_subagents=false)`** MCP tool + `GET /timeline` HTTP endpoint — single deduplicated, oldest-first cross-harness feed of every turn referencing an issue. Unions the structured field with an FTS keyword fallback (covering un-tagged turns), deduped by `doc_id`, sorted by `(timestamp, doc_id)`. +- **`get_issue_timeline(issue_id, limit=50, provider, date_from, date_to, include_subagents=false)`** MCP tool + `GET /timeline` HTTP endpoint: a single deduplicated, oldest-first cross-harness feed of every turn referencing an issue. Unions the structured field with an FTS keyword fallback (covering un-tagged turns), deduped by `doc_id`, sorted by `(timestamp, doc_id)`. - The timeline leaves out `agent-*` sessions (subagents and compactions) unless `include_subagents` is true. The output names the number of hidden turns. `/timeline` returns them as `hidden_subagents` and `hidden_subagents_capped`. - **Re-tagging stored rows:** `cleanup.py retag-issues` re-extracts issue IDs from every stored `document` with the active patterns and rewrites the changed rows in Milvus and FTS without re-embedding. It is a dry run by default. Run `--apply --yes` after a config change, with the server stopped. From 0c3d0b9959553594a37c44f6b5420b5786161b34 Mon Sep 17 00:00:00 2001 From: lbruton Date: Thu, 5 Nov 2026 12:05:49 -0600 Subject: [PATCH 3/3] chore: state that a capped hidden_subagents count is a lower bound --- DocVault/Overview.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/DocVault/Overview.md b/DocVault/Overview.md index 3534c9c..55fb741 100644 --- a/DocVault/Overview.md +++ b/DocVault/Overview.md @@ -95,7 +95,7 @@ Ingest scans turn text for tracker issue tokens. The deduplicated, uppercased se - **`issue_id` filter** on `search_all_sessions` / `search_session` — optional structured exact-token pre-filter (`issue_ids like "%,SESF-25,%"`), case-insensitive, combinable conjunctively with `provider` / `project_root` / `date_from`. Unset leaves search behavior byte-identical. - **`get_issue_timeline(issue_id, limit=50, provider, date_from, date_to, include_subagents=false)`** MCP tool + `GET /timeline` HTTP endpoint: a single deduplicated, oldest-first cross-harness feed of every turn referencing an issue. Unions the structured field with an FTS keyword fallback (covering un-tagged turns), deduped by `doc_id`, sorted by `(timestamp, doc_id)`. -- The timeline leaves out `agent-*` sessions (subagents and compactions) unless `include_subagents` is true. The output names the number of hidden turns. `/timeline` returns them as `hidden_subagents` and `hidden_subagents_capped`. +- The timeline leaves out `agent-*` sessions (subagents and compactions) unless `include_subagents` is true. The output names the number of hidden turns. `/timeline` returns them as `hidden_subagents` and `hidden_subagents_capped`. When `hidden_subagents_capped` is true, the FTS fallback filled its fetch window and `hidden_subagents` is a lower bound. - **Re-tagging stored rows:** `cleanup.py retag-issues` re-extracts issue IDs from every stored `document` with the active patterns and rewrites the changed rows in Milvus and FTS without re-embedding. It is a dry run by default. Run `--apply --yes` after a config change, with the server stopped. ## Specs