The Agent Skills Directory

[SAFE]: No malicious patterns detected. The skill instructions focus on process integrity, audit trails, and human oversight.
[PROMPT_INJECTION]: The skill processes untrusted data from external issue trackers which presents an injection surface, managed via procedural guardrails.
Ingestion points: Reads external issue bodies, comments, and labels as specified in SKILL.md.
Boundary markers: Relies on mandatory human-in-the-loop review and explicit state transitions rather than static delimiters.
Capability inventory: Writes to local ledger.yml and triage.md files; performs updates to external issue trackers (body, comments, labels) in SKILL.md.
Sanitization: Explicit instructions require human confirmation for all external actions and drift checks to prevent unauthorized data mutation or exfiltration.

gadd-triage