The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/image.mjs reads and copies authentication files (auth.json, config.toml) from the user's ~/.codex directory to a temporary folder to authenticate the codex CLI for image generation.
[DATA_EXFILTRATION]: Project briefs, which may contain sensitive internal requirements, are sent to the external domain https://mcp.seenry.design via scripts/research.mjs to retrieve design research data. This domain is not on the provided whitelist.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from a BRIEF.md file and interpolates it directly into persona-based prompts for design critique and motion judging in scripts/critic.mjs and scripts/choreography.mjs.
Ingestion points: BRIEF.md (read in scripts/check.mjs, scripts/critic.mjs, scripts/choreography.mjs).
Boundary markers: Minimal to none; the brief content is concatenated into instructions.
Capability inventory: File system access, execution of local binaries (spawn), and network requests (fetch).
Sanitization: No escaping or validation is performed on the ingested brief content.
[DYNAMIC_EXECUTION]: The skill relies on executing external system binaries. scripts/model_cli.mjs uses the which command to locate and spawn arbitrary codex or claude binaries found on the host system path.
[PERSISTENCE]: The script scripts/history.mjs creates and writes to ~/.config/seenry/history.json to maintain design identity state across multiple independent project sessions.

seenry