chore(agents): refresh the 21 audit agents and add terminal-fidelity-auditor - #393
Merged
Merged
Conversation
…auditor - Every agent brought in line with the current codebase (THI-353 ratchets, age-gate, support form); stale counts and paths removed. - Five agents that audit real data or access move to opus: route-attack, vercel-firewall, rbac-flow, classroom-workflow, user-forensics (13 opus / 8 sonnet). agentFrontmatter.test.ts now requires model opus|sonnet and a README entry for every agent. - Two secret-leak paths removed: no `curl -I` on protected URLs, no reads inside .secrets/. - A real user's identifier replaced by "utilisateur A" in an agent, STORY, ADR-009 and the B2B strategy doc. - New terminal-fidelity-auditor: runs lesson commands in the engine and in real GNU bash / PowerShell 7 inside a throwaway sandbox, and classifies each difference. - Theory replay extracted to src/test/lessonTheoryReplay.ts, shared by the test and the new `npm run theory:gaps` generator, which refuses to add a gap or raise BASH_SHOWN_ON_WINDOWS_MAX without --allow-new. - CLAUDE.md: Supabase migrations go through the Management API with the DevContext token; api-keys output is piped, never printed. Co-Authored-By: Claude Opus 5.5 <[email protected]>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
There was a problem hiding this comment.
Sorry @thierryvm, your pull request is larger than the review limit of 150,000 diff characters
This branch was successfully deployed
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Refresh of the 21 audit agents in
.claude/agents/, plus one new agent and a generator for the theory ratchet.lessonFidelity,lessonTheory), the age-gate, the support form. Stale counts and paths removed.route-attack-auditor,vercel-firewall-auditor,rbac-flow-tester,classroom-workflow-auditor,user-forensics-auditor. Result: 13 opus / 8 sonnet, no haiku.agentFrontmatter.test.tsnow requiresmodel: opus|sonnetand a README entry for every agent.curl -Ion protected URLs (the_vercel_jwtcookie carries the bypass token); no reads inside.secrets/.STORY.md, ADR-009 and the B2B strategy doc. Git history is not rewritten (decision recorded with @Thierry).terminal-fidelity-auditor. Runs lesson commands in the engine and in real GNU bash / PowerShell 7 inside a throwaway sandbox mirroringcreateInitialState, and classifies each difference. First run found two PowerShell gaps (Get-Locationheader,Get-Contentmissing-file message), queued for the next terminal PR.npm run theory:gaps. The theory replay moves tosrc/test/lessonTheoryReplay.ts, shared by the test and a new generator. The generator refuses to add a gap or raiseBASH_SHOWN_ON_WINDOWS_MAXwithout--allow-new.supabase projects api-keysoutput is piped togh secret set, never printed.Test plan
tsc --noEmit,eslint src,vitest run(2724 passed, 20 skipped; no.skip/.onlyadded by this diff),vite buildnpm run theory:gaps→162 -> 162 (removed 0, added 0), Windows count 37 → 37; an injected gap was refused (checked before this commit)feature-dev:code-reviewer: no finding at confidence ≥ 80; its one suggestion (the generated header is owned by the script) is applied🤖 Generated with Claude Code