The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes Python's subprocess module to execute various system commands including git, gh, docker, ps, and netstat for its auditing functions. Evidence: The LiveRunner class in scripts/harvest.py invokes subprocess.run with shell=False to collect system and repository information.
[INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves analyzing session transcripts which contain untrusted data from user turns and tool outputs. Ingestion points: scripts/harvest.py reads Claude Code transcript files from ~/.claude via the read_entries function. Boundary markers: The extraction and analysis logic does not implement specific delimiters or instructions to ignore potential commands embedded within the transcript text. Capability inventory: The skill has the capability to write to persistent files such as AGENTS.md and plans/*.md, and can execute system commands through the agent harness. Sanitization: While transcript data is loaded as JSON, the extracted text is processed without dedicated filtering or sanitization to mitigate prompt injection risks.
[EXTERNAL_DOWNLOADS]: The skill connects to remote repositories using git fetch and the GitHub CLI (gh) to retrieve updates and CI status. Evidence: SKILL.md documents network access for these tools to interact with repository remotes on well-known services.

session-harvest