The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository content (source code, documentation, workflow files) and utilizes the Bash tool, creating an attack surface for indirect prompt injection.
Ingestion points: The skill reads all tracked documentation and source files (manifests, scripts, schemas) within the repository during its audit and inventory phases (SKILL.md, Procedure Steps 1-2).
Boundary markers: The skill encourages the use of separate agent context files for session-specific rules, but does not implement specific delimiters or 'ignore' instructions for the repository content it processes (SKILL.md, reference/document-kinds.md).
Capability inventory: The skill uses Bash to execute git commands and arbitrary task steps during the 'newcomer test' (SKILL.md, Procedure Step 5).
Sanitization: No explicit sanitization or filtering is performed on repository content before it is interpolated into prompts or used to generate documentation drafts; however, the skill provides human-oriented safety guidance to avoid third-party writes during tests.
[EXTERNAL_DOWNLOADS]: The skill instructs the user or agent to install companion tools using the command npx skills add crissmoldovan/agent-skills. As these resources are owned by the skill's author ('crissmoldovan'), they are considered trusted vendor resources for this skill's context.
[DYNAMIC_EXECUTION]: The skill implements a 'newcomer test' protocol (reference/newcomer-test.md) which involves spawning a separate agent session to execute repository tasks in a clean environment. This is a form of dynamic task execution intended to verify that the documentation is sufficient for a new user.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to run git commands for baseline management, metadata retrieval (rev-parse, log), and file auditing (grep, glob).

layer-repository-docs