The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a local bash script scripts/gather_github.sh used to collect profile information from GitHub.
Evidence: The script executes gh api, gh repo list, and jq to process account metadata.
Mitigation: The instructions explicitly mandate that the agent must ask for user permission before running the script and wait for confirmation.
[EXTERNAL_DOWNLOADS]: The agent is instructed to fetch data from the GitHub API and user-provided external links (such as personal portfolios or LinkedIn profiles) to gather content for the README.
Evidence: Phase 2 and 3 of the workflow involve querying the GitHub API and fetching content from URLs provided by the user.
[PROMPT_INJECTION]: The skill processes untrusted external data (CVs, PDFs, and website content), which represents an indirect prompt injection attack surface.
Ingestion points: SKILL.md (Phase 3) describes reading user-uploaded CVs and fetching portfolio URLs.
Boundary markers: Not explicitly defined in the prompt instructions.
Capability inventory: Includes shell command execution via the gh CLI (scripts/gather_github.sh), file creation, and Git operations (commit/push).
Sanitization: The skill mitigates risks by instructing the agent to "verify every link," "never fabricate" facts, and seek explicit user confirmation before any repository writes.

github-profile-readme