Edmonds-Commerce-Limited/claude-code-hooks-daemon

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code

Single Source of Truth: An agent definition carries its role framing and role-unique behavioural instruction ONLY. It must not duplicate content from canonical docs — it links to them instead. No engineering-principle restatements, no code skeletons, no remediation cookbooks.

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code

Expert code review with zero tolerance for shortcuts. Analyzes real quality issues: dead code, confusion, TDD discipline, violations of engineering principles.

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code

Read-only documentation-SSoT auditor. Scans a project's documentation corpus (and, per its comment-hunt mandate, verbose source-code comment blocks that function as documentation) for conflicting truths, distributed/duplicated truths, stale derived facts and doc-shaped comments with no canonical home — the semantic…

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code

Before a new plan is filed, check whether an existing plan already covers the same work. Read-only; reports candidates with reasons, never a verdict.

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code

Implement Python code with strict adherence to project standards. Handles development, refactoring, and fixes using TDD, full type safety, and comprehensive testing.

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code needs its repo

Fix QA failures from qa-runner results. Resolves issues using proper fixes (never suppressions), maintains QA pattern documentation.

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code

Scope-deciding QA agent. Reads the change set (a branch/worktree or a range), picks the targeted checks per CLAUDE/QA.md, runs them and reports VERDICT, SCOPE, REASONING and FAILURES. Read-only; never fixes and never runs the full suite (a release step for the main thread).

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code

Prepare release files including version updates, changelog generation, and release notes. Validates prerequisites and prepares for Opus review. Does NOT commit or publish.

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code needs its repo

Specialist agent for diagnosing stop hook failures by reading Claude Code session transcripts. Use when the stop hook is not blocking as expected, Claude keeps stopping without required explanation, or preventedContinuation is false when it should be true.

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code

Quarantine execution space for safeguard-sensitive security work — defensive tasks whose source material a content-safety classifier may flag, which would silently substitute a different model for the rest of the caller's session. Runs on Opus so the caller's context never ingests that material. Delegate the WHOLE…

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code

Read-only security reviewer for this repository, dispatched one CHECK at a time by the security-review routines. Given a check id from CLAUDE/Routine/00001-security-review-full/CHECKS.md and an interval (or a diff), it hunts that one check's defect class and reports each finding with a file:line citation, the CLASS it…

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code

Review the Claude Code changelog between two versions and classify each change that touches what the daemon does as adopt / redundant / conflict / no impact. Read-only apart from its report file.

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code

Fact-check a plan (or a plan diff) against the current repository. Itemises every claim the text makes about the codebase, tries to disprove each with searches and reads, and reports VERIFIED / REFUTED / UNVERIFIABLE-HERE with evidence. Read-only apart from its report file.

Edmonds-Commerce-Limited/claude-code-hooks-daemon

Agent Claude Code

Runs the FULL QA gate (./scripts/qa/llmqa.py all) in the worktree it is given and reports VERDICT and failures to a file. The only agent type listed in exemptagenttypes; one full run at a time. Read-only; never fixes. Dispatch it with the worktree path and the report file path.