// Fix Python implementation issues identified by verify-command. Reads the verification report, analyzes discrepancies, applies fixes, and re-runs verification. Use when a verify-command report shows mismatches, or user says "fix report for XXX" or "patch XXX based on report".
Black-box usability testing from a user's perspective. Use when you want to test a module/function by writing code as a real user would — no peeking at source code, signatures, or docs. The tester only knows the basic import path and writes code purely from intuition and naming hints.
Learn and document a Stata command for Python econometric implementation reference. Use when user says "learn stata command XXX" or wants to understand a Stata command's internals for replicating in Python. Workflow: fetch help via Stata-MCP, write demo do-file, run it, and produce a command breakdown doc.
Cross-validate a Stata command against its Python implementation by running identical parameter combinations in both environments and comparing results. Use when user says "verify command XXX" or "validate XXX" or wants to check that Python output matches Stata output for a specific command.
| name | fix-from-report |
| description | Fix Python implementation issues identified by verify-command. Reads the verification report, analyzes discrepancies, applies fixes, and re-runs verification. Use when a verify-command report shows mismatches, or user says "fix report for XXX" or "patch XXX based on report". |
| agents | fix-analyzer fix-coder verify-runner verify-reporter |
| allowed-tools | Read Write Edit Grep Glob Bash(stata-mcp tool do .verify/codes/*) Bash(stata-mcp tool read-log *) Bash(uv run .verify/codes/*) Bash(uv run pytest tests/test_* -v) |
Read mismatches from .verify/reports/<command>.md, analyze root causes, fix Python code, then re-verify.
Read .verify/reports/<command>.md, confirm all ❌ items.
If report does not exist or all ✅ → stop, nothing to fix.
Use Agent tool to spawn fix-analyzer:
Agent(subagent_type="fix-analyzer", prompt="Read mismatches from .verify/reports/<command>.md, combine with docs/stata/command/<command>.md and src/ source code, analyze root causes and output fix suggestions to .verify/fixes/<command>.md")
Wait for fix-analyzer to complete the fix suggestion document.
Use Agent tool to spawn fix-coder:
Agent(subagent_type="fix-coder", prompt="Read fix suggestions from .verify/fixes/<command>.md, modify corresponding code in src/. Only change what needs changing, no extra refactoring.")
Wait for fix-coder to complete code modifications.
Spawn two agents in sequence:
Agent(subagent_type="verify-runner", prompt="Run .verify/codes/<command>.do and .verify/codes/<command>.py, collect raw output")
Agent(subagent_type="verify-reporter", prompt="Compare the following Stata and Python outputs item by item, generate new report overwriting .verify/reports/<command>.md")
uv run pytest tests/test_<command>.py -v
Ensure fixes did not break existing tests.
Summarize:
.verify/fixes/<command>.md.verify/reports/<command>.md (overwrites)stata-mcp is a CLI tool, not an MCP server: Verify with stata-mcp --version or uvx stata-mcp --version. All operations via stata-mcp tool <subcommand>python3 -c: Write script files then run. Also forbidden: python /dev/stdin <<, echo | python, etc.scipy.stats.t, scipy.stats.chi2, scipy.stats.f, etc.). All values must match Stata exactly, zero tolerance. Only acceptable difference is last-digit FP rounding drift, but should be minimized.## Fix Checklist section in reports is written by the verify workflow. The fix workflow must never modify it under any circumstances — no reading, checking, adding, deleting, or editing.> file, 2>&1, | tee, etc. (see verify-command skill's "Standard Execution")See references/ for a complete reghdfe example: