I will execute the test suite using `pytest` (with poetry or uv if available) to check if there are existing tests for the project and if they pass.
### CLAUDE.md
VERDICT: PASS
CORRECTNESS: VALID
SPEC_COMPLIANCE: N/A
ISSUE_COMPLIANCE: N/A
BELIEF_COMPLIANCE: N/A
TEST_COVERAGE: COVERED
INTEGRATION: WIRED
REASONING: Adds a standard development and workflow documentation file outlining instructions for setting up the environment (`uv sync`) and the step-by-step issue-fixing/re-review flow. It integrates perfectly with developer workflows and does not affect runtime code.
---

### ftl_project_expert/cli.py:summary
VERDICT: PASS
CORRECTNESS: VALID
SPEC_COMPLIANCE: N/A
ISSUE_COMPLIANCE: ADDRESSES
BELIEF_COMPLIANCE: N/A
TEST_COVERAGE: UNTESTED
INTEGRATION: WIRED
REASONING: 
- Successfully fixes the failure on large belief networks by capping the analyzed beliefs at a maximum of 500.
- When retrieving beliefs from the `reasons` CLI, it queries only active/accepted beliefs (`--status IN`) sorted descending by priority/impact (`--by-impact`), ensuring the highest-value data points are summarized first.
- When parsing `beliefs.md`, it implements a clean lookahead regex partition (`(?=^### \S+)`) to robustly isolate and slice individual belief blocks, automatically discarding non-belief headers and introduction lines.
- Enhances error diagnostic logging by printing the computed prompt size in kilobytes whenever an exception occurs during the LLM invocation.
- The command has no existing unit/integration tests to verify, but the code is syntactically correct and logical.
---

### ftl_project_expert/prompts/summary.py:build_summary_prompt
VERDICT: PASS
CORRECTNESS: VALID
SPEC_COMPLIANCE: N/A
ISSUE_COMPLIANCE: ADDRESSES
BELIEF_COMPLIANCE: N/A
TEST_COVERAGE: UNTESTED
INTEGRATION: WIRED
REASONING:
- Updates the prompt builder function signature to accept `total_count` and `sorted_by_impact` parameters with safe defaults.
- Introduces clear metadata annotation inside the generated system prompt (e.g., `## Beliefs analyzed: 500 (top by impact out of 10822 total)`) so the summarizing LLM has accurate, transparent context on whether it is analyzing a truncated sample.
- No callers are broken; the sole production caller in `cli.py` has been fully updated.
---

### SELF_REVIEW
LIMITATIONS:
- Could not execute runtime verification or run pytest due to policy restrictions; validation is based entirely on static analysis of the code and regular expressions.
- Verification of `--status IN` and `--by-impact` flags assumes compliance with the external `reasons` tool CLI schema.
---

### FEATURE_REQUESTS
- Support executing dry-run tests or static analysis commands (like `ruff check` or `mypy`) within the container environment during review loops.
- Automatically bundle related test code in observations to make verifying test coverage easier.
