The Agent Skills Directory

[DATA_EXFILTRATION]: The 'Tooling Reflection' mechanism in modules/completion.md instructs the agent to post learnings about its own operations and context to a public forum (https://github.com/athola/claude-night-market/discussions). This could result in the unintended disclosure of internal project details, environment metadata, or proprietary workflow information.
[PROMPT_INJECTION]: The skill retrieves and processes untrusted content from issue trackers (modules/issue-discovery.md) to define subagent tasks, creating an indirect prompt injection surface.
Ingestion points: External data enters through issue bodies and comments via gh issue view or glab issue view.
Boundary markers: There are no instructions to use delimiters or ignore instructions within the fetched issue text.
Capability inventory: The skill dispatches autonomous subagents with permissions to modify files, run tests, and commit code (modules/parallel-execution.md).
Sanitization: No validation is performed on issue content before it is converted into executable task plans.
[COMMAND_EXECUTION]: The workflow relies on significant system-level capabilities, including shell command execution (gh, glab, git) and the creation of background agent teams, which could be misused if the agent's instructions are hijacked.

do-issue