The Agent Skills Directory

[SAFE]: The skill is a project management and orchestration tool that demonstrates high security awareness. It includes explicit guardrails against prompt injection and unauthorized execution, maintaining a human-in-the-loop for strategic decisions. It follows best practices for repository management and governance.
[INDIRECT_PROMPT_INJECTION]: The skill acknowledges the risk of indirect prompt injection from GitHub issues and PRs. It establishes strict rules to treat external input as data rather than instructions, requiring sanitization and quoting of all external text. Findings are assessed as safe due to strong mitigation instructions.
Ingestion points: GitHub issues, PR descriptions, comments, and external SPEC files (SKILL.md, references/orchestrator-delegation-protocol.md).
Boundary markers: Explicit instructions to ignore embedded directives like 'ignore previous instructions' and to quote such text for user review instead of executing.
Capability inventory: GitHub CLI (gh), git, and delegation to other trusted internal skills like /execute-specs and /qa-analyst.
Sanitization: Instructions to sanitize/quote external text before use in commands.
[PROMPT_INJECTION]: Static analysis triggers for 'ignore previous instructions' are false positives. These phrases are found within the skill's own defensive instructions which command the agent to disregard such patterns if encountered in untrusted artifacts.
[EXTERNAL_DOWNLOADS]: The skill checks for updates to its own framework via 'git fetch'. It does not auto-update, but instead prompts the user to manually run the update command (npx skills add afonsoft/skills), ensuring human-in-the-loop control for external code execution. References to 'afonsoft/skills' are vendor-owned resources.

orchestrator