The Agent Skills Directory

[SAFE]: The skill incorporates strong data handling guidelines, explicitly instructing the agent to ensure that task descriptions and records do not contain credentials, customer data, or private URLs.
[SAFE]: Credentials management is handled securely; TOKEN-SETUP.md directs the user to input the Airtable Personal Access Token (PAT) directly into their local terminal rather than sharing it with the AI or storing it in project files.
[SAFE]: The skill uses localized tools (airtable-mcp, airtable-cli) and defined internal skills for its operations, with no evidence of unauthorized remote code execution or suspicious external downloads.
[SAFE]: Indirect prompt injection risk from Airtable task content is addressed through manual verification steps and specific constraints on the data allowed in the intake phase.
Ingestion points: server/workspaces/mirror/data/tasks.jsonl (SKILL.md) and live Airtable records (SKILL.md).
Boundary markers: None explicitly defined in prompt templates.
Capability inventory: Git push/commit commands, file-system writes, and airtable-cli mutations (SKILL.md).
Sanitization: Explicit instructions to exclude sensitive data, credentials, and customer information during the intake phase (SKILL.md).

airtable-flow