Agent Spec Sheet.

Single-agent specification: identity, purpose, inputs, outputs, tools, guardrails, escalation policy, governance, testing, and success metrics. One sheet per agent.

Filled during Redesign, refined during Build. Every agent shipped at Gate 4 has a current Spec Sheet.

HTML · This Page DOCX · Fillable All Templates

Faithful mirror This HTML is a read-only projection of lighthouse-agent-spec-sheet.docx. For filling in a live engagement, open the DOCX file.

§01

AGENT SPEC SHEET

Lighthouse Sprint — Stage 2: Redesign

Complete one spec sheet per agent. This document defines what the agent does, how it is triggered, what it integrates with, and how it is governed. It is the build contract — the Agent Engineer builds to this spec.

§02

1. Agent Identity

Agent ID	[AGT-01]
Agent Name	[Descriptive name]
Classification	[Fully Autonomous / Agent-Initiated Human-Approved / Human-Led Agent-Assisted]
Owner	[Agent Engineer name]
Blueprint Tasks	[Task #s from Redesign Blueprint this agent covers]
Version	[1.0]
Date	[YYYY-MM-DD]

§03

2. Purpose & Scope

What does this agent do? One paragraph. Be specific about the scope boundary — what it handles and what it explicitly does NOT handle.

[Write purpose statement here]

§04

3. Inputs

What data, signals, or triggers cause this agent to act? List every input with its source and format.

Trigger Event	[What initiates this agent? e.g., new RSS item, webhook, schedule, human request]
Input Data	[What data does it consume? List each source]
Input Format	[JSON, CSV, API response, email, etc.]
Input Source	[System/API/database that provides the data]
Prerequisites	[What must be true before this agent can run?]

§05

4. Outputs

What does this agent produce? Be explicit about format, destination, and success criteria.

Primary Output	[What is the main deliverable?]
Output Format	[Format and structure of the output]
Output Destination	[Where does the output go? System, queue, human review, etc.]
Side Effects	[Any secondary actions — notifications, logs, state changes]
Success Criteria	[How do you know the agent completed successfully?]

§06

5. Tools & Integrations

What systems, APIs, and tools does this agent use? List every external dependency.

System / API	Purpose	Auth Method	Failure Mode
[System]	[What for]	[API key / OAuth / etc.]	[What happens if unavailable]
[System]	[What for]	[API key / OAuth / etc.]	[What happens if unavailable]
[System]	[What for]	[API key / OAuth / etc.]	[What happens if unavailable]
[System]	[What for]	[API key / OAuth / etc.]	[What happens if unavailable]

§07

6. Guardrails & Constraints

What is this agent NOT allowed to do? Define hard limits, rate limits, and boundaries.

Hard Limits	[e.g., Cannot delete data, cannot exceed $X spend, cannot contact external parties]
Rate Limits	[Max invocations per hour/day, max API calls, max output volume]
Data Boundaries	[What data can it access? What is off-limits?]
Quality Thresholds	[Minimum accuracy, confidence score, or quality gate before output]
Timeout	[Max execution time before forced stop]

§08

7. Escalation Conditions

When should this agent stop and escalate to a human? List every condition.

Confidence Below	[Threshold — e.g., agent confidence < 80%]
Error Condition	[What errors trigger escalation?]
Edge Cases	[Scenarios the agent should not attempt to handle]
Escalation Target	[Who gets notified? Role, not person name]
Escalation Method	[Slack, email, queue, dashboard alert]

§09

8. Governance Integration

How does this agent integrate with the governance wrapper?

Monitoring	[What metrics are tracked? Latency, error rate, throughput, cost]
Alerting	[What triggers an alert? Thresholds and channels]
Audit Trail	[What is logged? Every invocation, inputs, outputs, decisions]
Override	[How can a human override or stop this agent mid-execution?]
Rollback	[How do you undo what this agent did if something goes wrong?]

§10

9. Testing Plan

How will this agent be validated before production deployment?

Unit Tests	[What specific behaviors are tested in isolation?]
Integration Tests	[What end-to-end flows are tested?]
Edge Case Tests	[What boundary conditions are tested?]
Load Test	[Expected volume and performance requirements]
Acceptance Criteria	[What must pass for Gate 3 approval?]

§11

10. Success Metrics

How will this agent's performance be measured against the baseline?

Baseline Metric	[Current performance from Baseline Metrics Sheet]
Target Metric	[Expected performance after deployment]
Measurement Method	[How will improvement be measured?]
Measurement Period	[Over what timeframe?]
Minimum Viable	[Lowest acceptable improvement to justify deployment]

Sign-off

Sprint Lead	Agent Engineer	Governance Architect
Name: ________________ Date: ________________	Name: ________________ Date: ________________	Name: ________________ Date: ________________