The Agent Skills Directory

[DATA_EXFILTRATION]: No unauthorized data exfiltration detected. Communication is limited to the official Trello API domain (api.trello.com) to perform stated functions.\n- [CREDENTIALS_UNSAFE]: Trello API Key and Token are collected via a setup script and stored locally in ~/.trello/config.json with restricted 600 permissions, following security best practices. No hardcoded secrets are present in the source code.\n- [COMMAND_EXECUTION]: Bash scripts wrap curl commands to interact with the Trello API. Parameters like card titles and descriptions are passed via command-line arguments. The agent is instructed to seek explicit user confirmation for additive or destructive actions.\n- [EXTERNAL_DOWNLOADS]: The skill relies on standard system package managers (brew, apt) for dependencies like jq and curl. No untrusted remote scripts are downloaded or executed.\n- [PROMPT_INJECTION]: The skill processes potentially untrusted data from Trello cards (titles, descriptions, comments) which represents a surface for indirect prompt injection. Ingestion points: scripts/trello-cards.sh (list, list-json, read, comments). Boundary markers: Absent. Capability inventory: scripts/trello-cards.sh (create, update, move, comment, archive, delete, position). Sanitization: Absent in scripts; risk is mitigated by agent instructions requiring human-in-the-loop confirmation.

trello