The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local CLI commands to interact with the repository and the openspec tool.
Evidence: Executes openspec status, openspec new change, openspec instructions proposal, openspec archive, and git log.
Context: These operations are necessary for initializing the onboarding environment and managing the lifecycle of project changes.
[PROMPT_INJECTION]: The skill processes untrusted data from the user's codebase and commit history, creating a surface for indirect prompt injection.
Ingestion points: Phase 2 (Codebase Analysis) scans files for comments like TODO or FIXME and reads the output of git log.
Boundary markers: The skill does not use explicit delimiters for the ingested code snippets in its internal instructions.
Capability inventory: The skill can write files and directories (Phase 4, 6, 7, 8) and execute local CLI tools (Phase 10).
Sanitization: The risk is mitigated by a mandatory human-in-the-loop (HITL) process where the agent drafts suggestions, proposals, and task lists for user review and explicit approval before proceeding with any implementation or filesystem changes.

openspec-onboard