Use when defining testable product requirements for an agent's task scope, autonomy, budgets, intervention, success, and irreversible-action gates; use ai-agent-strategy-doc for portfolio strategy.
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Use when defining testable product requirements for an agent's task scope, autonomy, budgets, intervention, success, and irreversible-action gates; use ai-agent-strategy-doc for portfolio strategy.
defining testable product requirements for an agent's task scope, autonomy, budgets, intervention, success, and irreversible-action gates; use ai-agent-strategy-doc for portfolio strategy.
Use this procedure when the required source artefacts are available and AI agent feature PRD specification is the next lifecycle deliverable.
Do Not Use When
Use ai-agent-strategy-doc when that neighbouring route owns the decision or deliverable.
Do not invent missing project evidence, standards clauses, thresholds, or stakeholder decisions.
Required Inputs
Artefact
Source or provider
Required?
Behaviour when missing
Approved agent strategy, user journeys, action catalogue, and risk limits
Product owner, operator, and AI governance owner
Yes
Stop the affected step, name the missing source, and return only a qualified gap record.
Workflow
Inspect the required inputs and log the exact sources, versions, and unresolved assumptions.
Apply this skill's existing domain workflow and decision rules to produce AI agent feature PRD specification.
Stop when a required source, accountable decision owner, or deterministic test oracle is absent.
Recover by preserving valid work, marking the blocked scope, and returning the narrowest qualified artefact plus the next evidence needed.
Outputs
Artefact
Consumer
Acceptance condition
AI agent feature PRD specification
Agent architecture, evaluation, red-team, and operations teams
Required sections are populated, source links resolve, and every material requirement or decision has an observable review or test oracle.
Evidence Produced
Evidence
Reviewer
Acceptance condition
Source, decision, trace, and validation record for AI agent feature PRD specification
Requirements quality reviewer
Inputs used, decisions made, checks run, failures, and unassessed items are explicit.
Capability and permission boundaries
Read and search are required. Editing is allowed only when the request authorises creation or repair of the named requirements artefact. Publishing, production mutation, destructive action, spending, and certification require explicit authority.
Degraded mode
Fallback: if a required file, reviewer, standard source, network check, renderer, or execution capability is unavailable, return the narrowest useful qualified result and mark the affected check not assessed; never convert an unassessed check into a pass.
Decision Rules
Choice or condition
Action
Failure or risk avoided
An action is irreversible or high impact
Require explicit approval, evidence capture, and a tested recovery path.
Unbounded agent side effects.
Required inputs and test oracles are complete
Continue through the existing workflow and record evidence.
A deliverable whose acceptance cannot be reproduced.
A mandatory source or owner is missing
Stop the affected branch and issue a qualified gap record.
Fabricated context or unauthorised decisions.
Quality Standards
Preserve stable identifiers and bidirectional traceability from project evidence to AI agent feature PRD specification and its acceptance checks.
Apply ISO/IEEE measures only with a named metric, method, threshold, evidence source, and responsible reviewer; run the anti-slop gate before release.
Anti-Patterns
Producing AI agent feature PRD specification from assumed context. Fix: cite the project source or mark the scope blocked.
Accepting a material requirement without a deterministic oracle. Fix: add a measurable result, boundary, and verification method.
Crossing into ai-agent-strategy-doc without routing the decision. Fix: hand off the named input and preserve trace links.
Treating an unavailable check as passed. Fix: mark it not assessed and state the release consequence.
Claiming standards, statutory, or stakeholder approval without evidence. Fix: cite the source and reviewer or qualify the claim.
Produces the agent-feature complement to ai-feature-prd-spec. Every agent FR carries seven mandatory agent clauses that the AI feature PRD does not collect. The acceptance gates point at the agent eval rig and the agent red-team plan.
What is shown at the approval moment (the plan, the diff, the action arguments).
What undo / revert is available after the fact.
How a user contests an action that already executed.
Reference EU AI Act Art. 14 (human oversight) where the feature is high-risk.
Step 5: Define rollout posture per agent FR
Initial rollout for every agent FR begins in shadow mode (agent proposes; human acts). Promotion stages: shadow → canary at L1 → L2 → L3 if applicable. Reference the agent rollout runbook for stage gates.
Step 6: Define acceptance tests against the agent eval rig
Every agent FR has a row in the agent eval rig with:
Golden-task set ID.
Replay-set ID.
Adversarial-set ID (links to the agent red-team plan).
Pass thresholds per metric.
CI gate definition.
Step 7: Write the spec
AI_Agent_Feature_PRD_Spec.md sections: 1) Agent FR Inventory, 2) Per-FR Agent Clauses, 3) Success Metrics, 4) Human-in-the-Loop Placement, 5) Rollout Posture, 6) Eval & Red-Team Acceptance Gates, 7) Traceability to PRD, AI Feature PRD, and to agent eval / red-team IDs.
Standards
IEEE 830-1998
NIST AI RMF MAP / MEASURE
EU AI Act Art. 13 (transparency), Art. 14 (human oversight)
