The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions direct the agent to fetch and adopt behavioral guidance from external command outputs (openspec instructions archive and openspec instructions specs).\n
Ingestion points: Identified in SKILL.md (Steps 1 and 4), where the agent runs CLI commands to retrieve JSON payloads containing context, operationGuidance, and rules from the local environment.\n
Boundary markers: The skill attempts to define boundaries by stating these inputs are "advisory" and should not override "controlling inputs," yet it also explicitly labels context as a "required prompt-level input" and rules as constraints on spec writing.\n
Capability inventory: The agent has permissions to execute Bash commands, read project files (like tasks.md), and write to the filesystem via the spec synchronization process.\n
Sanitization: There is no automated validation or sanitization of the externally provided instructions; the skill relies on the agent's ability to identify conflicts manually.\n- [COMMAND_EXECUTION]: The skill frequently uses the Bash tool to interact with the openspec CLI and standard filesystem utilities.\n
Evidence: Execution of openspec status, openspec list, mkdir -p, and mv commands to manage project state and archive change directories.

openspec-archive-change