Plan Lite

Plan Lite

Turn a sufficiently clear outcome into the smallest plan that makes execution, verification, review, and rollback legible.

Read [the local v2 contract](references/usable-loop-v2.md) for complete record fields and authority/invalidation rules.

Preconditions

Use the current request, Alignment Snapshot, or Outcome Contract as the outcome source. An unconflicted current user request is authoritative when the user owns the task. Agent-derived interpretation is provisional. If a missing decision would materially change the plan, expose it instead of burying it as a task.

Do not require a spec. Do not plan a clear one-step edit. Do not execute the plan or silently invoke another handrail.

Build the implementation map

State:

1. goal, acceptance criteria, and non-goals;
2. chosen implementation strategy and major tradeoffs;
3. ordered vertical slices with observable completion;
4. verification at the narrow, integration, and risk-specific levels;
5. isolation/rollback needs and explicit authorization gates;
6. known dependencies, unknowns, and stopping conditions.

Prefer independently verifiable slices over file-by-file activity lists. Name likely files only when they clarify navigation; planned paths are hypotheses.

Review Boundary Seed

Produce a compact seed containing:

• exact Outcome Reference binding;
• expected logical changes;
• dependent/generated changes;
• authorization guardrails;
• known pre-existing changes;
• participating workspaces;
• base/recovery hints;
• open boundary questions;
• Path semantics: hypotheses-not-allowlist.

The seed does not approve omitted paths and does not classify future changes. Branch Sandbox may record actual task-start delta from it. Two-Axis Review or Verify Before Claim must inventory and classify current candidates mechanically.

Direct result

Return the strategy, ordered slices, verification matrix, main risk/rollback, and Review Boundary Seed. Keep it conversational unless the user asks for a durable artifact. Bind Task ID, Outcome Revision, and Authority in one line by default; do not expand the full Outcome Reference or record schema unless the user explicitly asks for it or continuity requires it. Use a Checkpoint only on a continuity trigger.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Aquish-Lee
• Source: Aquish-Lee/agent-handrails
• License: MIT

Install and usage instructions live in the source repository linked above.