The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a series of openspec CLI commands (e.g., openspec new change, openspec status, openspec instructions) to manage the planning lifecycle. These interactions are restricted by the platform's allowed-tools configuration to the openspec namespace, ensuring command execution remains within the authorized scope of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill operates on external data by reading project files and instructions generated by the CLI. It identifies this surface and provides clear instructions to the agent to treat this data as background constraints rather than executable commands or raw content for inclusion in outputs. Evidence:
Ingestion points: Reads JSON outputs from openspec instructions and markdown content from project dependency artifacts identified in SKILL.md.
Boundary markers: Includes a 'Planning boundary' section that explicitly restricts the agent from implementing changes or editing project code, limiting the impact of potential injections.
Capability inventory: Limited to reading/writing planning artifacts within the openspec environment; no arbitrary network or system write capabilities are present.
Sanitization: Instructions mandate the use of specific output templates and explicitly forbid the copying of raw <context> or <rules> blocks from the input data into the final files.
[DYNAMIC_EXECUTION]: The skill allows for the invocation of specific sub-commands or skills if the openspec tool's instruction field delegates a task. This dynamic behavior is safely encapsulated within the tool's defined planning workflow and constrained by the same shell access limits that apply to the primary commands.

openspec-propose