The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external project files and CLI outputs that can influence agent behavior.
Ingestion points: Reads openspec/config.yaml (specifically the context field) and the output of openspec instructions (including context, rules, and instruction fields).
Boundary markers: The instructions include a "Planning boundary" section to explicitly restrict the agent from implementation or editing project code.
Capability inventory: The skill uses Bash(openspec:*) to execute local commands and has permissions to write artifacts to the filesystem.
Sanitization: There is no explicit sanitization or structured validation of the content retrieved from the CLI output before it is used as instructions.
[DYNAMIC_EXECUTION]: The skill supports executing commands determined at runtime by an external source.
In Step 6a, the skill is instructed to invoke a "specific skill or command" if the instruction field from the CLI response delegates creation to one. This allows the tool's output to dynamically control the agent's tool invocation.
[COMMAND_EXECUTION]: The skill relies on shell command execution for its primary functionality.
It utilizes the openspec CLI via Bash to manage changes, resolve paths, and fetch metadata (openspec new change, openspec context, openspec status, openspec instructions). The usage is scoped to the openspec toolset via the allowed-tools configuration.

openspec-propose