The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently constructs and executes shell commands by interpolating variables such as change names and directory paths.
Evidence: Step 1, 2, 4, and 5 use interpolated strings to call openspec CLI commands and standard utilities like mkdir and mv.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources and instructs the agent to treat them as 'behavior contracts' or 'operation guidance'.
Ingestion points: tasks.md (Step 3), openspec instructions archive (Step 1), and openspec instructions specs (Step 4).
Boundary markers: The instructions (Step 1 and 4) include explicit boundaries, directing the agent to keep external inputs separate from built-in steps and resolved paths, and to report conflicts rather than blindly following instructions.
Capability inventory: The skill has the ability to write to the file system (mkdir, mv, openspec sync) and execute CLI tools.
Sanitization: The workflow checks for valid JSON formatting and command exit statuses, but does not specify further content sanitization for the ingested instructions.
[DYNAMIC_EXECUTION]: The skill invokes an inline sub-workflow (openspec-sync-specs) to perform merges, which involves dynamic processing and writing of specification files.
Evidence: Step 4 details the invocation of an 'agent-driven intelligent merge' and verification of resulting files.

openspec-archive-change