The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves fetching and analyzing untrusted content from public repositories (SKILL.md files, source code, CI configurations) which could contain malicious instructions or deceptive content.
Ingestion points: The scripts/fetch_repo_assets.sh script programmatically fetches files from external repositories into the docs/research/ directory.
Boundary markers: The references/attribution-rules.md and references/apply-protocol.md files establish boundaries by requiring manual user approval and explicit attribution for all extracted insights.
Capability inventory: The skill utilizes gh CLI, git, curl, and jq for data retrieval and processing, with the ability to write to the local filesystem.
Sanitization: The instructions specifically mandate that the agent must rewrite all extracted patterns in a 'local voice' rather than copying verbatim, acting as a manual sanitization step for ingested text.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch repository data, metadata, and security signals from several well-known developer services.
Evidence: The scripts and documentation reference interactions with github.com, api.scorecard.dev (OpenSSF Scorecard), ossinsight.io, api.deps.dev (Google OS Insights), libraries.io, sourcegraph.com, and gitlab.com.
Trust Scope: These services are recognized as established developer infrastructure and are used for their intended research purposes.
[COMMAND_EXECUTION]: The skill relies on executing system commands to perform its research tasks.
Evidence: The scripts/ directory contains bash scripts that invoke gh, git, jq, curl, and base64 as subprocesses to search for repos, fetch assets, and perform history forensics.

research-git