The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data that could influence its behavior, specifically through instructions fetched at runtime.
Ingestion points: Step 2 and Step 7 read JSON output from openspec instructions, which contains context, operationGuidance, and rules fields intended to guide the agent.
Boundary markers: The skill includes explicit instructions for the agent to treat these fields as advisory, separate them from controlling inputs (conflict analysis, CLI contracts), and preserve controlling values in case of conflict.
Capability inventory: The skill has the capability to execute shell commands (via the openspec CLI), create directories (mkdir), and move project files (mv).
Sanitization: Instructions require the agent to explain why any guidance is rejected and prohibit the verbatim copying of these external inputs into the codebase or summaries.
[COMMAND_EXECUTION]: The skill executes shell commands to perform file system operations and manage project state.
Evidence: The skill uses openspec list, openspec status, and openspec instructions to read state, and uses mkdir -p and mv to archive files.
Constraints: Tool usage is scoped in the frontmatter to Bash(openspec:*), though the markdown body also includes standard system commands like mkdir and mv for directory management.

openspec-bulk-archive-change