The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface due to its core functionality of fetching external content. \n
Ingestion points: Untrusted data is retrieved from external issue trackers (GitHub, Linear, Jira) and arbitrary URLs during Phase 0 (Triage). \n
Boundary markers: Although the skill wraps internal arguments in tags, it lacks explicit markers or instructions to isolate data fetched from remote issue comments, which could potentially contain malicious instructions. \n
Capability inventory: The agent has access to file system modifications, dependency installation commands (npm, bun, bundle), and arbitrary test execution capabilities. \n
Sanitization: There is no specified logic for sanitizing or filtering instructions that might be embedded within the retrieved issue descriptions or comments.\n- [COMMAND_EXECUTION]: The skill involves the execution of various system and development tools. \n
Evidence: Instructions explicitly mention running npm install, bun install, bundle install, git log, git bisect, and specialized tools like /ce-commit-push-pr. \n
Context: These commands are standard for a technical debugging environment and are necessary for the skill's functionality.\n- [EXTERNAL_DOWNLOADS]: The skill retrieves data from remote repositories and tracking services. \n
Evidence: Uses the GitHub CLI (gh issue view) and web fetching to pull issue titles, bodies, and comment threads. \n
Context: This behavior is transparently described and required to provide the triage and context needed for automated debugging.

ce-debug