The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to discover the skill's installation directory (find ~/.claude ~/.config ~/.agents ...) and to run a bundled Python linter (python3 assets/postmortem_lint.py). These operations are restricted to specific configuration directories and bundled scripts, and are used to manage local assets and verify report quality.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze untrusted external data sources, including git history, CI logs, and issue tracker threads, which introduces a surface for indirect prompt injection if those sources contain malicious instructions.
Ingestion points: The skill reads git log, git blame, CI runs, app logs, and issue/PR threads to reconstruct failure timelines (as described in SKILL.md).
Boundary markers: The instructions do not define explicit boundary markers or warnings for the agent to ignore embedded instructions within these data sources.
Capability inventory: The skill has the capability to read file system artifacts and run a bundled Python linter script (assets/postmortem_lint.py).
Sanitization: There is no specific logic described to sanitize or filter the content of ingested logs or comments before they are processed by the agent.
[SAFE]: The Python scripts provided with the skill (assets/postmortem_lint.py, eval/run_eval.py) are offline-first, use only the standard library, and do not perform network operations.

bug-autopsy