from aiskillstore
Execute a README-first smoke test and produce standardized reproducibility outputs (`repro_outputs/`) and PATCHES.md — trusted reporting for repo reproduction r
Provides a focused execution-and-reporting workflow for README-first AI repos. The skill runs a specified smoke test or documented inference/evaluation command, captures execution evidence, normalizes outputs into a repro_outputs/ directory, and writes patch notes (PATCHES.md) when repository files change during the run. It is intentionally narrow: it does execution/evidence and standardized reporting, not full environment setup or long-running training.
Use after a reproduction target and environment plan already exist and you need a trusted, auditable record of a single run: smoke tests, evaluation commands, or short inference checks. Avoid using it for initial repo scanning, heavy training runs, or when the target is undecided.
scripts/run_command.py and scripts/write_outputs.py (has_scripts=true)references/reporting-policy.md and example reporting templates (has_references=true)repro_outputs/, PATCHES.md).Designed for CI-style automation and developer assistants that perform reproducibility work (agents that can run commands, collect artifacts, and write structured reports).
Skill provides README-first smoke test execution and standardized repro output reporting for AI repos. run_command.py is well-written with proper argparse, timeout handling, and git status diffing, but requires --repo and --command args (fails gracefully). write_outputs.py has a brittle path resolution (parents[3]) that crashes outside the original repo structure. SKILL.md is lean and well-structured with clear scope boundaries.
No security concerns. The subprocess execution is by design (the skill's purpose is to run commands). shlex.split provides proper argument splitting. write_outputs.py's hard-coded relative path traversal (parents[3]) is a code quality issue, not a security one.
