The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository being audited, including manifests, configuration files, and source code, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: SKILL.md Step 1 (Passive Environmental Inspection) and Step 5 (Execute Migration) involve reading arbitrary files from the target repository to diagnose and refactor the codebase.
Boundary markers: The skill implements a 'Project loading boundary' for cross-skill references and uses a mandatory 'Approval Gate' (Step 3) for all proposed actions. It also uses a specific 'User decision format' defined in references/portable-practice.md to ensure human review, though it does not explicitly sanitize the code contents being processed.
Capability inventory: The skill is authorized to perform Git operations (SKILL.md Step 4), modify files via AI refactoring (SKILL.md Step 5), install packages and run validation commands (SKILL.md Step 6), and commit changes (SKILL.md Step 7).
Sanitization: Relies on human-in-the-loop review and explicit user approval for all proposed modifications and command executions to filter potential injection attempts.
[DYNAMIC_EXECUTION]: The skill generates new source code and configuration files at runtime and subsequently executes them via validation gates.
Evidence: SKILL.md Step 5 (Execute Migration) instructs the agent to 'translate complex configurations, adapt deprecated API calls, and update mocks/tests'. This AI-generated content is then executed or validated in Step 6 (Automated Validation Gates) using commands like npm install, pytest, or build scripts.

repo-modernizer-acrazie