The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, creating a potential surface for indirect prompt injection. Ingestion points: Data enters the agent's context through JSON outputs from the openspec CLI (status, instructions) and by reading project artifacts from disk. Boundary markers: The skill explicitly defines a 'Planning boundary' and uses a structured JSON schema for instructions. Capability inventory: The agent can execute openspec commands via Bash and write artifact files to resolved paths. Sanitization: The skill instructs the agent to sanitize user-provided change names into kebab-case but relies on the CLI for instruction validity.
[DYNAMIC_EXECUTION]: The workflow allows the openspec tool to specify a command or skill in its instruction field for the agent to invoke at runtime. This dynamic delegation is an intended feature of the OpenSpec ecosystem for managing complex artifact generation steps.
[METADATA_POISONING]: The skill's internal author field is 'openspec', while the distribution context identifies the author as 'full-stack-skills'. This minor metadata discrepancy represents an inconsistency in identifying information but does not appear to indicate malicious intent.

openspec-propose