The Agent Skills Directory

[PROMPT_INJECTION]: The skill-from-codebase functionality is susceptible to indirect prompt injection because it derives instructions for the AI agent directly from scanning the repository codebase and git history.
Ingestion points: Scans the full repository and processes git history from last_generated_commit..HEAD as specified in skill-from-codebase/SKILL.md.
Boundary markers: No specific markers or delimiters are defined to isolate data-derived content from instruction-based content in the generated SKILL.md.
Capability inventory: The skill performs file system writes at the repository root and executes shell-based git commands.
Sanitization: While it utilizes a 'Validator Subagent' for quality gates like broken links, there is no explicit sanitization logic to detect or neutralize malicious instructions embedded in the scanned source code.
[COMMAND_EXECUTION]: The skill requires the execution of shell commands to perform repository analysis.
Evidence: skill-from-codebase/SKILL.md mandates the use of git status --porcelain and git diff to determine impact analysis and gate the documentation refresh process.

nexus