The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands using node:child_process (execSync) within the collect-evidence.mjs script. These commands are used to interact with the local environment and the GitHub CLI (gh). Specific operations include retrieving the repository root (git rev-parse --show-toplevel), identifying the current branch and commit SHA, and checking repository visibility (gh repo view). The instructions also guide the agent to perform git push and gh pr create operations after user confirmation.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it processes external data from test transcripts and Playwright reports.
Ingestion points: The scripts/collect-evidence.mjs script reads files provided via the --transcript flag or discovered in playwright-report and test-results directories.
Boundary markers: The script wraps ingested content in fenced code blocks within the generated PR_BODY.md. It uses alternating delimiters (~~~) if the content itself contains triple backticks to prevent markdown escaping.
Capability inventory: The skill possesses file read/write capabilities and the ability to execute Git and GitHub CLI commands.
Sanitization: The skill includes a proactive security feature in scripts/collect-evidence.mjs that scans ingested text files for high-signal secret patterns (e.g., AWS keys, Slack tokens, GitHub tokens, and Bearer tokens) using the SECRET_PATTERNS regex list. It alerts the user if sensitive data is detected before the evidence is committed or pushed.

safe-pr