Merge pull request #27 from stranma/fix/remove-ci-review

fix: Remove redundant CI-based Claude Code review workflow

Author: stranma

.github/workflows/claude-code-review.yml

@@ -163,7 +163,7 @@ actual scope at completion based on workspace signals.
 
 - **5 workflow agents** -- code quality, test coverage, PR writing, code review, docs updates
 - **3 security hooks** -- block destructive commands, scan for leaked secrets, catch Unicode injection
-- **CI/CD** -- GitHub Actions for lint + test + typecheck + publish + AI code review
+- **CI/CD** -- GitHub Actions for lint + test + typecheck + publish
 - **Tool stack** -- [uv](https://docs.astral.sh/uv/) workspaces, [ruff](https://docs.astral.sh/ruff/), [pyright](https://github.com/microsoft/pyright), [pytest](https://pytest.org/), [hatchling](https://hatch.pypa.io/)
 
 ### Optional specialists

README.md

@@ -360,7 +360,7 @@ These run automatically as part of `/done` for Standard and Project scope tasks.
 
 **What.** Reads the diff and applies the review rules (architecture, code-quality, performance, test). Reports findings with severity levels. Read-only (`permissionMode: dontAsk`, `memory: project` for CLAUDE.md context).
 
-**Remove.** No pre-PR code review. The GitHub Actions `claude-code-review.yml` workflow still runs an AI review on the PR itself, so review is delayed but not eliminated.
+**Remove.** No pre-PR code review. Linters and type checkers still catch mechanical issues, but logic errors, security concerns, and design problems would only be caught during manual review.
 
 </details>
 
@@ -614,17 +614,6 @@ All workflows are in `.github/workflows/`. They use `{{base_branch}}` placeholde
 
 </details>
 
-<details>
-<summary>claude-code-review.yml -- AI Code Review</summary>
-
-**Why.** Automated code review on PRs using Claude. Catches issues that linters cannot: logic errors, security concerns, design problems.
-
-**What.** Triggers on PR open/sync/ready. Uses `anthropics/claude-code-action@v1` with Sonnet. Reviews for correctness, security, performance, readability, and testing. Has read-only tools (Read, Glob, Grep -- no Bash). Posts findings as PR comments. 10-minute timeout.
-
-**Remove.** No AI review on PRs. The local `code-reviewer` agent (run by `/done`) still provides pre-PR review.
-
-</details>
-
 <details>
 <summary>template-integration.yml -- Template Validation</summary>
 

docs/ARCHITECTURE_GUIDE.md

@@ -24,7 +24,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - 3 slash commands (`/cove`, `/cove-isolated`, `/security-audit`) provide chain-of-verification for accuracy and a 6-phase security posture scan with A-F grading
 - 3 new specialized agents: `security-auditor` (OWASP-based vulnerability analysis, read-only), `refactoring-specialist` (SOLID/code smell detection, read-only), `output-evaluator` (LLM-as-Judge quality scoring for automated pipelines)
 - 4 review rules in `.claude/rules/` auto-loaded as project context -- cover architecture, code quality, performance, and test quality concerns that linters cannot catch
-- AI-powered PR review via GitHub Actions (`claude-code-review.yml`) using `anthropics/claude-code-action@v1` -- automatically reviews PRs with read-only tools on open/sync/ready_for_review
 - Local config template (`.claude/settings.local.json.example`) for developer-specific hook overrides, extra permissions, and MCP server configuration
 - Tests for hooks (test_hooks.py), commands (test_commands.py), agents (test_agents.py), and rules (test_rules.py) validating existence, structure, frontmatter, and behavioral contracts
 - New `docs/DEVELOPMENT_PROCESS.md` contains the full Q/S/P development workflow, agent reference, failure protocol, changelog format, and PCC shorthand -- accessible on demand instead of always loaded into context
@@ -57,6 +56,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Removed
 
+- CI-based AI code review workflow (`claude-code-review.yml`) -- the local `code-reviewer` agent (run by `/done` at step S.6.5) already provides equivalent pre-PR review coverage, making the CI workflow redundant and removing the need to manage an `ANTHROPIC_API_KEY` secret in GitHub
 - `/catchup` command -- its context restoration role overlaps with `/sync`, which already covers pre-flight workspace state
 - `/ship` slash command -- its 3-tier validation checklist (Blockers, High Priority, Recommended) is preserved in `/done` Phase 2
 - Shell Command Style and Allowed Operations sections from CLAUDE.md -- absolute path preferences and read-only command lists are now handled by settings.json permission rules rather than prose instructions

docs/CHANGELOG.md

@@ -77,7 +77,7 @@ When a decision is superseded or obsolete, delete it (git history preserves the
 - output-evaluator uses haiku + dontAsk -- designed for automated pipeline integration, scoring is formulaic
 - Review rules have no `paths:` frontmatter (apply globally) and stay under 80 lines -- loaded into every context window
 - CLAUDE.md kept compact per ETH Zurich paper decision; detailed hooks/commands/rules tables added to DEVELOPMENT_PROCESS.md instead
-- CI review workflow uses claude-sonnet-4-6 with read-only tools (Read, Glob, Grep) -- security principle of least privilege
+- ~~CI review workflow~~ removed -- see 2026-03-13 decision below
 
 ## 2026-03-02: QSP Enforcement and Pre-flight Sync
 
@@ -172,3 +172,12 @@ When a decision is superseded or obsolete, delete it (git history preserves the
 | GitHub API via curl (`curl -H "Authorization: ..." https://api.github.com/.../merge`) | Blocking curl to github.com is fragile and breaks legitimate web fetching. The hook already blocks commands containing `GH_TOKEN=` as a literal argument. | Use fine-grained PATs with minimal scopes. CLAUDE.md instructs Claude to use `gh` CLI, not raw API calls. Token scoping is the real control. |
 | Docker not present but deny rules exist | Docker is not installed in the current template container. Deny rules exist as defense-in-depth for users who add Docker-in-Docker later. | If Docker-in-Docker is added, the deny list should be revisited (add `-v` and `--mount` volume escape patterns). |
 | Whitelisted domains as exfil channels | `github.com` is whitelisted for git/gh operations. A compromised agent could theoretically exfiltrate via gist creation or issue comments. | Token scoping (no gist/issue create permission) + GH mutation deny rules in Tier 2. Tier 3 accepts this risk explicitly. |
+
+## 2026-03-13: Remove CI-Based Claude Code Review
+
+**Request**: Remove the `claude-code-review.yml` GitHub Actions workflow. The local `code-reviewer` agent (run by `/done` at step S.6.5) already provides equivalent pre-PR review coverage, making the CI workflow redundant.
+
+**Decisions**:
+- Delete `claude-code-review.yml` entirely -- the local code-reviewer agent provides the same review before PR creation, and the CI workflow required managing an `ANTHROPIC_API_KEY` secret in GitHub
+- Keep `dangerous-actions-blocker.sh` `ANTHROPIC_API_KEY=` pattern unchanged -- it blocks secrets in commands generally, not CI-specific
+- Keep `docs/IMPLEMENTATION_PLAN.md` unchanged -- historical record of completed work

docs/DECISIONS.md

@@ -121,7 +121,7 @@ See [Devcontainer Permissions](docs/DEVCONTAINER_PERMISSIONS.md) for the full de
 - **CLAUDE.md** -- compact agent directives (~40 lines) with /sync, /design, /done workflow
 - **5 workflow agents** -- code quality, test coverage, PR writing, code review, docs updates
 - **3 security hooks** -- block destructive commands, scan for leaked secrets, catch Unicode injection
-- **CI/CD** -- GitHub Actions for lint + test + typecheck + publish + AI code review
+- **CI/CD** -- GitHub Actions for lint + test + typecheck + publish
 - **Tool stack** -- [uv](https://docs.astral.sh/uv/) workspaces, [ruff](https://docs.astral.sh/ruff/), [pyright](https://github.com/microsoft/pyright), [pytest](https://pytest.org/), [hatchling](https://hatch.pypa.io/)
 
 ### Optional specialists