Conversation
- Add py-agent skill and script (format, lint, typecheck, test) - Auto-detect runner (uv, poetry, plain) and python binary (python3/python) - Auto-detect tools: ruff/black (format), ruff/flake8 (lint), mypy/pyright (typecheck), pytest/unittest (test) - CI-aware: format checks in CI, auto-fixes locally; MAX_LINES unlimited in CI - Support --fail-fast flag (passes -x to pytest) - Fix hints after each failure with auto-fix suggestions where available - Add clean and issues scenario test fixtures - Update README, install.sh with py-agent entry
In fix mode, run --check first to detect which files need formatting, then apply the fix and list the changed files in the output. Ported from ai-barometer's cargo-agent.
![ai-barometer-staging[bot]](https://avatars.githubusercontent.com/u/205648601?s=60&v=4){kind=small}
![cadence[bot]](https://avatars.githubusercontent.com/in/2857168?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cadence Session Review
| Score |  Solid execution with minor scope gaps and one ordering bug |
| Models | Claude (opus-4.6) |
| Sessions | 5 |
| Phases | 60% implementation, 20% cross-cutting refactors, 15% debugging, 5% documentation |
Model delivered py-agent plus cross-cutting improvements (fail-fast, Fix: hints, CI-aware MAX_LINES) to all existing agents in a single sweep. Followed output contract conventions well, including scenario fixtures.
- py-agent script is well-structured, mirrors cargo/npm/terra patterns faithfully
- Cross-cutting refactors (fail-fast, Fix: hints, CI MAX_LINES) applied consistently across all agents
- CI workflow not updated to include py-agent scenarios — scope gap from session ordering
--helpbroken outside Python project dirs due to check ordering- Session included good iterative debugging (inverted logic fix, python3 fallback)
Recommendations
Prompting — Enumerate all cross-cutting files in the prompt
When adding a new agent that should follow existing patterns (including CI), explicitly mention all artifacts that need updating. The model applied changes to docs and install.sh but missed ci.yml because the CI workflow was created in a separate session.
Before
The CI workflow was created in an earlier session, and the py-agent session never referenced updating it.
Reframe
"Add py-agent with full workflow. Also update ci.yml to add a py-agent job matching the existing patterns."
Tip
Keep a checklist of cross-cutting files (ci.yml, install.sh, README.md) and name them in the prompt.
Agent instructions — Add help-accessibility rule to agent conventions
The model placed the Python project existence check before the help case, breaking --help outside a project. Adding an explicit rule in AGENTS.md about help accessibility would prevent this pattern across all agents.
Reframe
Add to AGENTS.md: - The help/usage command must work without any project context (no project detection before help).
Prompting — Request edge-case testing for CLI flags
When verifying CLI behavior, prompt the model to test edge cases like running from outside a valid project directory. The model only tested happy paths from within fixture dirs.
Before
The model only tested --help from within the fixture directory.
Reframe
After running --help, also run: cd /tmp && py-agent.sh --help to verify help works outside a project directory.
Tip
A good pattern: 'Test the command from both inside and outside a valid project directory.'
- Add py-agent CI job with baseline and with-ruff matrix variants - Add workflow-level flock (Linux) / Perl fallback (macOS) to prevent concurrent py-agent runs - Add CHANGED_FILES support: scopes format and lint to changed .py files; typecheck and test run project-wide (need full context) - Add CHANGED_FILES to SKILL.md allowed-tools and env knobs table
- Move help/--help before project-existence checks (py-agent + convention) - Switch have_tool() to use `which` via runner instead of --version - Make unittest fallback conditional on python being available - Add SKIP result when no test runner is found - Update definition-of-done with help-accessibility requirement
1 participant
Summary
py-agent— lean Python workflow runner (format, lint, typecheck, test)--fail-fast,Fix:hints, and all shared env knobsTest plan
--helpoutput verifiedtests/run-scenarios.sh py-agent