The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon dynamic instructions provided by external commands, creating a surface for indirect prompt injection.
- Ingestion points: The skill executes openspec instructions archive (Step 1) and openspec instructions specs (Step 4), parsing their JSON output for context, operationGuidance, and rules fields.
- Boundary markers: The instructions explicitly tell the agent to "read and consider" and "follow entries" from the external output. While it includes guardrails to not copy text verbatim and to report conflicts with built-in logic, the agent is still directed to incorporate this external data into its reasoning process.
- Capability inventory: The skill has the capability to execute openspec commands via Bash, read local files (e.g., tasks.md), and perform directory operations (mkdir, mv).
- Sanitization: The skill specifies that guidance is advisory and should not be used to infer replacement paths or be copied verbatim into specs or summaries. It also requires the agent to explain if guidance is ignored due to conflicts.
[COMMAND_EXECUTION]: The skill constructs shell commands (mkdir -p and mv) using variables (planningHome.changesDir, changeRoot, target-name) derived from the JSON output of the openspec status command. While the skill correctly uses double quotes to wrap these variables, it relies on the integrity of the data returned by the openspec tool to avoid potential path traversal or argument injection issues during the archiving process.

openspec-archive-change