The Agent Skills Directory

[DATA_EXFILTRATION]: The skill implements a 'Fork Pipeline' designed to move content from a private working copy to a public GitHub fork. While it references a 'sanitize-rules.json' for pruning private content, the automated nature of pushing potentially sensitive internal source code to a public forum presents a risk of data exposure.
[PROMPT_INJECTION]: The skill processes untrusted input from the community during 'piper search' (GitHub issues) and 'piper triage' (PR review). The lack of explicit boundary markers and the agent's access to powerful tools like 'Bash' create a surface for indirect prompt injection where malicious external content could influence agent actions.
Ingestion points: piper search, piper triage (reads GitHub issue and PR content).
Boundary markers: Absent.
Capability inventory: Bash, Write, Edit, git push, gh issue create (defined in SKILL.md).
Sanitization: Absent for incoming untrusted data; only mentioned for outgoing code via sanitize-rules.json.
[COMMAND_EXECUTION]: The skill relies extensively on the 'Bash' tool to execute GitHub CLI (gh) operations, Git commands (subtree push, repo sync), and local automation scripts.
[REMOTE_CODE_EXECUTION]: The skill executes local Python scripts (fork_pipeline.py, task_queue.py) using python3 -m to handle sensitive logic such as project building and 'sanitize scans.'
[EXTERNAL_DOWNLOADS]: The sync-upstream command uses gh repo sync to fetch and integrate code from an external repository (hoangsonww/Claude-Code-Agent-Monitor) into the local environment.

piper