The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the local system through the openspec CLI. The execution environment is restricted to this specific toolset via the allowed-tools configuration, which limits the potential impact of command execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data including user-provided descriptions, project files, and dynamic instructions returned by the CLI.
Ingestion points: Reads project background, rules, and templates from the openspec instructions CLI command output and re-reads dependency files from the disk.
Boundary markers: The skill contains explicit instructions to treat CLI-provided context and rules as internal constraints rather than content for the output files, which acts as a logical delimiter for the agent.
Capability inventory: Limited to writing Markdown files and executing openspec commands.
Sanitization: The skill mandates a 'planning boundary' where the agent is prohibited from editing project code or starting implementation, regardless of the input instructions.

openspec-propose