The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external files and tool outputs which could contain adversarial instructions designed to influence agent behavior.
Ingestion points: Reads content from tasks.md (Step 3) and JSON payloads from openspec instructions archive and openspec instructions specs (Steps 1 and 4).
Boundary markers: The instructions explicitly mandate that operationGuidance and context are advisory and must not override "controlling inputs" such as CLI contracts, resolved paths, or explicit user choices. It requires the agent to report and explain any conflicts.
Capability inventory: The skill possesses the ability to execute openspec CLI commands and manipulate the file system via mkdir and mv commands.
Sanitization: Relies on structural JSON parsing of CLI outputs and logical verification of synced specs before finalizing the archive.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform system-level operations.
Shell Operations: Executes mkdir -p and mv to manage the archive directory structure.
Variable Interpolation: Commands are constructed using variables like <changeRoot> and <target-name> derived from tool outputs or user input. While intended for local repository management, these represent a standard injection surface if the executing agent does not apply strict shell escaping.
CLI Invocation: Uses the openspec CLI for gathering project status, listing changes, and fetching workflow instructions.

openspec-archive-change