Skip to content

[pull] main from Alishahryar1:main - #250

Merged
pull[bot] merged 2 commits into
Co-Contribute-OpenSource:mainfrom
Alishahryar1:main
Sep 27, 2026
Merged

pull[bot] merged 2 commits into
Co-Contribute-OpenSource:mainfrom
Alishahryar1:main

Conversation

@pull

@pull pull Bot commented Sep 27, 2026 •

Copy link
Copy Markdown

See Commits and Changes for more details.


Created by pull[bot] (v2.0.0-alpha.4)

Can you help keep this open source service alive? 💖 Please sponsor : )

Alishahryar1 and others added 2 commits September 26, 2026 20:29
## Why

Let users select FCC models directly in VS Code's built-in Chat and
Agent mode. Native clients also need a consistent fallback when
providers omit context metadata, while keeping output allowances and
reasoning effort distinct.

## How

Add a VS Code Chat integration card with Connect, Disconnect, an
actionable Retry after update failures, and a modal showing the
configuration path. Reconcile failures and recovery even when polling
misses the updating phase or a status read is already in flight, without
replaying reload notices. Generate its Custom Endpoint model group from
FCC's application catalog and reconcile connected configurations when
models or settings change.

Use a shared 200,000-token context fallback at client serialization
boundaries. Direct model discovery includes whether context came from
the provider or the default, so launcher decoding preserves unknown
metadata and OpenCode retains its existing output-reserve policy. Apply
the fallback to supported Codex, VS Code, Pi, OpenCode, DSH, Aider,
Cline, Hermes and Muse catalog fields.

Expose VS Code's native effort picker using shared Codex effort labels
and the existing FCC reasoning pipeline. Prefer reported output limits,
otherwise allocate up to 81,920 tokens, capped at half the context to
leave room for input. These are output ceilings, not effort-dependent
thinking budgets.

Accepted configuration contracts:

- Target the standard VS Code default profile on Windows, macOS and
Linux. Users select their own chat and utility models.
- FCC owns the generated model inventory and capabilities. Refresh
replaces manual edits inside generated model definitions. Preserve
native effort preferences in `group.settings`, the group name, unrelated
group properties and other providers' groups.
- Use Messages and explicit generated models. Store FCC's managed local
token in supported per-model request headers. VS Code treats the
group-level `apiKey` as a secret-store reference.
- Use private POSIX file permissions on replacement and repair
permissions on unchanged refreshes. Permission-only repairs do not
produce settings-reload notices. Windows retains the existing
user-profile directory ACL policy.
- Preserve JSON values rather than comments or formatting. FCC
serializes its own writes and replaces files atomically. External
editors do not participate in its lock.
- Reported context wins over the fallback. Native clients without a
supported per-model context setting, including the Claude protocol
clients, keep their native behavior. Other harnesses retain their
existing output and effort policies.

Correct the browser-shutdown test to coordinate readiness under its
existing parent-process deadline. Isolate the real thread-diagnostic
integration test in a bounded subprocess with stage output. The previous
macOS blocking operation remains unconfirmed, so this supplies
diagnostic evidence without changing the helper speculatively.

Use synchronous request-arrival counters in browser tests and drain the
VS Code retry test's forwarding callbacks before context disposal. Give
server-fixture shutdown one thirty-second watchdog, report thread and
async-task diagnostics after five seconds, and stop subsequent tests if
the owned server remains alive. This accommodates the existing
native-interrupt grace period while preserving evidence of the
unexplained Windows SQLite delay.

Validation: 6,453 pytest tests and 221 Playwright tests passed locally,
alongside Ruff and ty. Native VS Code 1.139.1 with NVIDIA NIM sent
low/high/none effort and the configured output cap, created a file, and
completed a follow-up edit. Cross-platform CI will validate the pushed
commit, including the macOS diagnostic boundary.












<!-- greptile_comment -->

<!-- greptile_summary -->

<h2><a
href="https://app.greptile.com/api/retrigger?id=70031025"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://greptile-static-assets.s3.amazonaws.com/badges/RetriggerDark.svg?v=2"><source
media="(prefers-color-scheme: light)"
srcset="https://greptile-static-assets.s3.amazonaws.com/badges/Retrigger.svg?v=2"><img
alt="Retrigger"
src="https://greptile-static-assets.s3.amazonaws.com/badges/Retrigger.svg?v=2"
align="right"></picture></a>Confidence Score: 5/5</h2>

<!-- greptile-risk -->

No blocking issue remains. The outstanding shutdown-cleanup concern is
non-blocking.

<details open><summary>Summary</summary>

The PR adds native VS Code Chat integration and a shared context
fallback for supported clients. The latest change adds VS Code Chat to
the README’s coding-agent list. No new actionable issue was found in
that change.
</details>

<!-- greptile_confidence_score:5 -->

<sub>Reviews (7) · Last reviewed commit: ["Count VS Code Chat as the
eleventh
codin..."](https://github.com/alishahryar1/free-claude-code/commit/9d8eb5f7c5ecadc265bf1ce0b8ef50be8c28a03c)</sub>

<!-- /greptile_comment -->
## Why

Add OrcaRouter as a built-in cloud provider so users can configure its
API key and discover its models through FCC.

Closes #1751.

## How

Use the shared OpenAI-compatible transport with OrcaRouter's fixed
endpoint, named reasoning effort, credential validation, and model
discovery. Filter out models explicitly marked as incompatible with Chat
Completions.

Include the provider logo and website link in the standard
alphabetically sorted cards and configuration modal. Add the README
provider row and update the total to 56 providers.

---------

Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Alishahryar1 <20476625+Alishahryar1@users.noreply.github.com>
@pull pull Bot locked and limited conversation to collaborators Sep 27, 2026
@pull pull Bot added the ⤵️ pull label Sep 27, 2026
@pull
pull Bot merged commit 10fb1d0 into Co-Contribute-OpenSource:main Sep 27, 2026
10 of 11 checks passed

This branch had an error being deployed

1 failed deployment
pypi — 10fb1d02 Deployed Sep 27, 2026 by pull[bot] via Publish to PyPI #7
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants