Skip to content

TEMPLATE.md

Latest commit

 

History

History
50 lines (30 loc) · 2.09 KB

TEMPLATE.md

File metadata and controls

50 lines (30 loc) · 2.09 KB

[AGENT NAME] — Operational Spec

IDENTITY

Designation: Brand (if client-facing): Function: [one sentence — what does this agent do?]

[Write 2-3 sentences defining who this agent is, how it communicates, and what it is NOT. Be specific. "You are a helpful assistant" is not an identity. "You are a technical writer who produces API documentation in plain English for developer audiences" is.]

PLATFORMS

[List every platform, service, or channel this agent interacts with. Include relevant API details, authentication notes, and any platform-specific constraints.]

  • Platform 1: [role, access details, constraints]
  • Platform 2: [role, access details, constraints]

TRIGGER

[How does this agent get activated? Options: manual (someone fires it), cron (scheduled), event (responds to a signal). Be explicit about what triggers a run and what doesn't.]

WORKFLOW

[Step-by-step sequence from trigger to deliverable. Number each step. Include decision points — what does the agent do when something is ambiguous? When should it proceed vs. flag for human review?]

  1. [First action after trigger]
  2. [Next step]
  3. [Decision point — if X, do Y; if uncertain, flag]
  4. [Build/draft/execute the deliverable]
  5. [Self-check — specific criteria, not "review your work"]
  6. [Output location and handoff — where does the deliverable go, who reviews it?]

OUTPUT STANDARDS

[What does good output look like? Be specific. Include: format, tone, length constraints, quality bars. If you can't describe what bad output looks like, your standards aren't specific enough.]

TOOL PERMISSIONS

[Exactly which tools this agent is allowed to use. Anything not listed is implicitly denied.]

  • [tool 1 — what it's used for]
  • [tool 2 — what it's used for]

STANDING LIMITS

[Hard boundaries. What this agent must NEVER do, regardless of context. Publishing, spending, client interaction, scope creep — whatever applies to your operation.]

TOKEN DISCIPLINE

[How should this agent manage its own resource consumption? When should it stop iterating and flag instead? What's the ceiling before it asks for help?]