M1: Assistant config + skill feature flag enforcement#10152
Merged
noanflaherty merged 2 commits intoFeb 27, 2026
Merged
Conversation
Co-Authored-By: Claude <noreply@anthropic.com>
noanflaherty
requested review from
awlevin and
siddseethepalli
as code owners
Contributor
Author
👀 Where to focus your review
Risk level: Medium — New enforcement layer that silently removes skills from multiple surfaces. Misconfigured flags would make skills invisible with no error in the UI list or model prompt. |
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
…t section Add isSkillFeatureEnabled checks in skill_load for child skills in both the body-loading loop and the loaded_skill marker loop, so flag-OFF child skills are fully hidden. Also filter hardcoded browser/twitter references in buildDynamicSkillWorkflowSection through isSkillFeatureEnabled so the system prompt does not advertise disabled skills. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Contributor
Author
Review feedback addressedPushed fix commit Issue 1: Enforce feature flags on included child skills in
|
Contributor
Author
Contributor
Author
noanflaherty
merged commit fcdbfd1
into
feature/skill-feature-flags-gateway
2 checks passed
|
Codex Review: Didn't find any major issues. Nice work! ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
7 tasks
noanflaherty
added a commit
that referenced
this pull request
Feb 27, 2026
* M1: Assistant config + skill feature flag enforcement (#10152) * feat: add skill feature flags config and enforcement Co-Authored-By: Claude <noreply@anthropic.com> * fix: enforce feature flags on included child skills and dynamic prompt section Add isSkillFeatureEnabled checks in skill_load for child skills in both the body-loading loop and the loaded_skill marker loop, so flag-OFF child skills are fully hidden. Also filter hardcoded browser/twitter references in buildDynamicSkillWorkflowSection through isSkillFeatureEnabled so the system prompt does not advertise disabled skills. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com> * M2: Gateway feature-flags API + persistence (#10165) * feat: add gateway feature-flags REST API with config persistence Co-Authored-By: Claude <noreply@anthropic.com> * fix: handle malformed URI encoding and config parse errors in feature-flags - Wrap decodeURIComponent in try/catch in gateway/src/index.ts to return 400 Bad Request on malformed percent-encoding instead of crashing to the global error handler with a misleading 500. - Refactor readConfigFile to use a discriminated union result type that distinguishes "file doesn't exist" (returns empty config) from "file exists but can't be parsed" (returns error). The PATCH handler now returns 500 with a descriptive message when the config file is malformed, preventing silent data loss. The GET handler gracefully degrades to empty flags on parse errors (no data loss risk). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: validate parsed config is a plain object in readConfigFile --------- Co-authored-by: Claude <noreply@anthropic.com> * M3: Client-only PATCH auth split for feature flags (#10171) * feat: add client-only PATCH auth split for feature-flags Co-Authored-By: Claude <noreply@anthropic.com> * fix: remove rate-limit on 403 and handle identical token edge case - Remove authRateLimiter.recordFailure() from 403 path to avoid penalizing legitimately authenticated clients who used the wrong token type (Issue 1) - Only record failures on 401 (truly invalid authentication) - Skip runtime-token rejection when FEATURE_FLAG_TOKEN is identical to runtimeBearerToken to support single-token deployments (Issue 2) Addresses review feedback on PR #10171 --------- Co-authored-by: Claude <noreply@anthropic.com> * M4: Client token plumbing (macOS/iOS) (#10175) * feat: add feature-flag client token plumbing for macOS/iOS Co-Authored-By: Claude <noreply@anthropic.com> * fix: throw on unavailable transport, always use gateway on macOS, clear stale token Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com> * docs: add feature-flags architecture documentation (#10181) Co-authored-by: Claude <noreply@anthropic.com> * fix: use active transport for macOS remote mode, preserve token on re-pair Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: address review feedback on final PR #10183 1. [P1] Fix gateway type-check: remove dead `not_found` variant from ConfigReadResult union (fixes narrowing bug on `result.detail`), make `featureFlagToken` optional in GatewayConfig type. 2. [P1] Add featureFlagToken to pairing responses: runtime now reads ~/.vellum/feature-flag-token and includes it in approval responses so iOS receives the token during QR pairing. 3. [P1] Fix PATCH auth bypass when tokens equal: loadConfig() now detects token collision and regenerates the feature-flag token to ensure the auth split is always enforceable. 4. [P2] Gate guardian-verification section on feature flags: system prompt skips buildGuardianVerificationRoutingSection() when guardian-verify-setup skill flag is OFF. 5. [P2] Fix failing integration test: removed incorrect assertion that <available_skills> disappears when only browser/twitter flags are OFF (bundled skills remain visible). 6. [P2] Fix macOS local HTTP transport: setFeatureFlag now checks if httpTransport.baseURL is localhost before delegating — local HTTP mode (localHttpEnabled) falls through to the direct gateway call. 7. [P2] Accept dotted skill IDs in feature-flag keys: updated regex from [a-z0-9_-]+ to [a-z0-9][a-z0-9._-]* to match managed skill ID validation. * Migrate hatch visibility to skill flag and remove deprecated client flags --------- Co-authored-by: Claude <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add featureFlags config section to assistant and enforce feature flags at all skill exposure code paths. When a skill flag is OFF, the skill is hidden from UI lists, model prompts, skill_load, and active tool projection. Part of #10145.