The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to override standard agent safety protocols by implementing a 'No-Stall Principle'. It directs the AI to skip all user-confirmation gates and suppress checkpoints that would normally allow for human oversight, particularly at sub-skill returns and loop boundaries.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted issue descriptions from a remote repository to perform automated code modifications.
Ingestion points: Issue content (titles and bodies) is ingested from the 'SonicGarden/dev-workflow-issues' repository via the 'gh issue list' tool (SKILL.md Step 2).
Boundary markers: The skill lacks explicit boundary markers or delimiters to isolate untrusted issue data from the agent's internal logic.
Capability inventory: The skill has permissions to perform 'Edit' and 'Write' operations on local files, and execute 'git commit' and 'git push' to remote repositories (SKILL.md Step 3.4, 3.7, 4).
Sanitization: There is no evidence of sanitization or validation of the ingested data before it is used to influence code changes.
[COMMAND_EXECUTION]: The skill performs extensive shell operations using 'git' and 'gh' (GitHub CLI) to manage branches, modify repository history, and interact with external issues. These commands are executed autonomously as part of the triage workflow.
[DATA_EXFILTRATION]: The skill is designed to automatically push code changes to a remote repository ('git push') and post comments to GitHub issues ('gh issue comment') based on decisions derived from external issue content. This creates an automated channel for data and code to be sent to external infrastructure without manual verification.

dev-workflow-triage