The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches external context and operation guidance via the openspec instructions archive and openspec instructions specs commands. These instructions are intended to influence the agent's behavior during the archiving and syncing processes, creating a surface where attacker-controlled data could manipulate agent logic. \n
Ingestion points: Step 1 and Step 4 in SKILL.md execute commands that return JSON payloads containing instructions for the agent. \n
Boundary markers: The skill includes natural language instructions to prioritize 'controlling inputs' over external guidance, but these are not enforced by technical delimiters. \n
Capability inventory: The skill performs file system modifications (mv, mkdir) and executes various openspec CLI commands. \n
Sanitization: There is no evidence of programmatic sanitization or schema validation for the data returned by the instruction commands. \n- [COMMAND_EXECUTION]: The skill leverages the Bash tool to interact with the openspec CLI and perform file system management tasks. \n
Evidence: The skill uses commands such as openspec list, openspec status, mkdir -p, and mv to process and relocate project change directories.

openspec-archive-change