The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches external instructions using openspec instructions archive and openspec instructions specs. These instructions, including context, operationGuidance, and rules, are explicitly incorporated into the agent's reasoning process. If these are sourced from untrusted changes, they could attempt to influence the agent's behavior during the archiving or syncing process.
Ingestion points: Output from the openspec instructions archive (Step 1) and openspec instructions specs (Step 4) tools.
Boundary markers: The skill provides logical instructions to handle conflicts (e.g., 'If context conflicts... report the conflict') and warns against verbatim copying, but does not specify technical delimiters to isolate this untrusted content from the primary prompt.
Capability inventory: Access to bash for executing openspec, mkdir, and mv commands, as well as file system read access for tasks.md and specification files.
Sanitization: The skill mandates the agent to perform a conflict analysis between external guidance and built-in rules, rejecting guidance that contradicts controlling inputs.
[COMMAND_EXECUTION]: The skill executes shell commands (mkdir, mv) to manage the archive directory. These operations depend on variables such as planningHome.changesDir and changeRoot which are derived from the output of the openspec status tool. While these are standard operations for archiving, they represent a capability to modify the filesystem structure based on dynamic tool output.

openspec-archive-change