The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data including source code, documentation, and review reports which represent a potential surface for indirect prompt injection.
Ingestion points: Data enters the context via the recovery of tasks in SKILL.md and the interpretation of reports as described in references/review-signals.md.
Boundary markers: The skill relies on standard Markdown headers and YAML frontmatter delimiters to structure instructions and separate metadata from content.
Capability inventory: The skill utilizes git commands for repository state and the intent CLI for file management, metadata synchronization, and validation.
Sanitization: The instructions in references/review-signals.md explicitly direct the agent to treat external reports and linked content as untrusted and prohibit the execution of commands or access to secrets based on that input.
[EXTERNAL_DOWNLOADS]: The skill references the use of npx skills and gh skill to fetch and install agent skills from remote repositories.
Evidence: references/distribution.md details how consumers can install skills using these tools.
Context: These operations refer to established tools within the developer ecosystem, and the instructions emphasize security best practices like pinning to specific commit SHAs rather than mutable tags.
[COMMAND_EXECUTION]: The skill requires the execution of the intent CLI tool and various git commands for task validation and repository maintenance.
Evidence: SKILL.md and references/maintainer-commands.md provide specific command patterns for validation and synchronization (e.g., npm exec intent validate).
Context: The instructions mandate using locally installed versions of these tools in devDependencies to ensure predictable behavior and avoid runtime fetch risks.

generate-skill