The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository (code, documentation), issue trackers (issue bodies, PR descriptions), and existing instruction files to generate guidelines.
Ingestion points: Processes content from the repository root, package manifests, READMEs, issue trackers, and pull request data (defined in SKILL.md and references/codebase-discovery.md).
Boundary markers: Implements an explicit "Untrusted content boundary" that instructs the agent to treat external directives (e.g., "ignore previous instructions") as data rather than instructions, and to report them as suspected injections (references/agentic-setup.md).
Capability inventory: The skill can read repository files, perform shell-based verification (grep, wc), and write new markdown files to the local file system after user confirmation.
Sanitization: Enforces credential redaction, path validation using regular expressions (^[A-Za-z0-9._/-]+$), and prohibits the execution of commands that require credentials or affect external state (references/agentic-setup.md, references/codebase-discovery.md).
[COMMAND_EXECUTION]: The skill utilizes shell commands for repository discovery and file verification.
Verification steps: Uses grep and wc to validate the structure, headings, and size of generated AGENTS.md files (references/verification.md).
Validation commands: Suggests the execution of repository-specific commands (e.g., tests, linters) discovered in manifests, but mandates user confirmation and explicitly excludes dangerous operations like deployment or database migrations (references/agentic-setup.md).

om-create-agents-md