The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides structured instructions to a subagent to perform technical analysis. There are no attempts to bypass safety filters or override agent constraints.
[DATA_EXFILTRATION]: The skill reads project-level configuration files and source code structure. It utilizes the Gemini CLI for semantic analysis, which is a well-known service. No sensitive credentials or private data are sent to unauthorized external domains.
[COMMAND_EXECUTION]: The skill uses the Bash tool to interact with the local 'ccw' CLI, manage project files in .workflow/ and .ccw/, and gather project metadata via git. These actions are strictly aligned with the skill's purpose of project initialization.
[REMOTE_CODE_EXECUTION]: No remote script downloads or executions from untrusted sources were found. The skill relies on local environment tools.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The subagent reads project manifest files (e.g., package.json, requirements.txt, go.mod) to determine the tech stack.
Boundary markers: Structural markers are used to separate task assignments from analysis data.
Capability inventory: The skill possesses file system access (Read, Write) and command execution (Bash, spawn_agent).
Sanitization: The results of the technical analysis are reviewed and confirmed by the user through a multi-round interactive questionnaire (ASK_USER), providing a human-in-the-loop verification step.

spec-setup