The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied context and external evidence to build project roadmaps. It mitigates injection risks by instructing the agent to treat all retrieved information as 'evidence, not authorization' (SKILL.md) and providing a specific evidence chain requirement (references/artifact-paths.md) that involves validating JSON keys and types without returning raw contents.
[COMMAND_EXECUTION]: The skill instructions allow for the execution of standard Git commands (git ls-files, git check-ignore) to manage repository state and ensure that internal metadata in .tigerkit/ is properly ignored by version control. These operations are scoped to artifact management and do not provide a general shell execution surface.
[EXTERNAL_DOWNLOADS]: The skill references external methodology resources on GitHub (phuryn/pm-skills and mattpocock/skills) for provenance and documentation purposes. These references are pinned to specific commit hashes, representing a secure development practice for tracking upstream sources.
[SAFE]: The skill defines robust security practices for handling temporary files and secrets, including using restrictive file permissions (0700/0600) for secret inputs and requiring immediate deletion of credentials after consumption (references/artifact-paths.md).

tk-roadmap