The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository files to generate AGENTS.md documentation intended to guide AI agents. Malicious instructions in source code could be incorporated into these files, poisoning the agent's context. Ingestion points: Phase 2 scanners read all tracked project files to extract facts. Boundary markers: The generated templates lack boundary markers or warnings to ignore embedded content. Capability inventory: Uses Write/Edit tools for file modification and executes shell commands for environment setup. Sanitization: The skill does not perform sanitization of extracted facts before writing them to the knowledge base files.
[COMMAND_EXECUTION]: Phase 5 executes a sequence of shell commands to generate project snapshots and manage local git configuration. Evidence: Use of git ls-files, xargs, wc -l, sed, and find to calculate project metrics and modify the .git/info/exclude file.
[DYNAMIC_EXECUTION]: The skill employs dynamic execution for repository analysis and metadata generation. Evidence: Usage of eval cells for repository sizing calculations in Phase 1 and node -e for inline script execution during the snapshot phase.

init-deep