The Agent Skills Directory

[COMMAND_EXECUTION]: The script prior_work.py executes ripgrep (rg) and git commands via subprocess.run to perform local file searches and verify repository states. These are standard operations for the skill's discovery functionality.
[PERSISTENCE]: The skill includes an installation command (--install in prior_work_hook.py) that modifies agent configuration files such as ~/.claude/settings.json and ~/.codex/hooks.json. This is intended to register the skill's hooks into the agent's lifecycle across sessions.
[DYNAMIC_EXECUTION]: The skill supports the execution of external commands defined in a local manifest (sources.json) to act as search adapters. Additionally, it uses importlib to dynamically load its own internal hook module for auditing and classification purposes.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources (files and command outputs) during retrieval. While this presents an attack surface where malicious instructions could be embedded in the content being searched, the skill uses these results to populate a structured 'reuse receipt' and includes logic in its hooks to classify and validate prompts, mitigating accidental obedience.

prior-work-retrieval