The Agent Skills Directory

[SAFE]: The skill implements exceptionally defensive programming patterns to ensure secure operation within a repository. It requires explicit user authorization for any out-of-scope or destructive actions and prohibits the autonomous installation of new tools.
[INDIRECT_PROMPT_INJECTION]: The skill manages the risk of ingesting malicious instructions from external sources through multi-layered validation.
Ingestion points: The skill reads task data from GitHub, Jira, and Linear, as well as local specification artifacts (references/intake.md).
Boundary markers: Explicit instructions in SKILL.md command the agent to treat tracker and fetched content as untrusted evidence that cannot override the skill's own rules. Furthermore, references/approved-spec.md mandates SHA-256 hash validation for canonical artifacts to prevent tampering.
Capability inventory: The skill has the capability to modify repository files and execute shell commands for verification, but these are strictly gated by the requirement for verified spec artifacts or explicit user instructions.
Sanitization: The skill requires redacting sensitive tokens, cookies, and private payloads from all generated receipts and evidence logs (references/evidence.md).
[EXTERNAL_DOWNLOADS]: The skill includes documentation links in references/provenance.md to repositories from well-known sources (Matt Pocock and Cursor/Anysphere). These links are provided for attribution and to document the provenance of the skill's logic; they do not trigger runtime code downloads or installations.

implement