The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data (markdown/TSV files and user-provided issue descriptions) and possesses capabilities to execute shell commands and perform network operations via the GitHub CLI.
Ingestion points: Markdown/TSV report files (processed by scripts/render-report.py) and user input for GitHub issues (processed by scripts/ps-issue.sh).
Capability inventory: Command execution via gh CLI (network/write) and python3 script execution.
Boundary markers: Instructions require the agent to propose actions and obtain human confirmation.
Sanitization: scripts/ps-issue.sh implements a redact function to strip common API keys, tokens, and PII from issue bodies before submission.
[COMMAND_EXECUTION]: The skill utilizes external CLI tools and local scripts to perform its functions.
Evidence: Invocation of gh issue list, gh issue create, and python3 scripts/render-report.py.
[DATA_EXFILTRATION]: The skill provides a mechanism to transmit information to a remote repository for issue reporting.
Evidence: The scripts/ps-issue.sh script facilitates sending data to the pedroknigge/pstack-skills GitHub repository.
Mitigation: The implementation includes a mandatory human-in-the-loop (HITL) confirmation step (--confirm) and automated redaction of sensitive credentials.

pstack