Executing Plans

Executing Plans

Overview

Load plan, review critically, execute all tasks, report when complete.

Announce at start: "I'm using the executing-plans skill to implement this plan."

For non-trivial execution, include Aegis Visibility: briefly tie the active slice to its plan, checkpoint, drift or verification boundary. At completion, pass plan adherence, evidence, complexity and residual risk to verification-before-completion for the unified receipt.

If subagents are available and the plan has genuinely independent tasks, prefer subagent-driven-development; lack of subagent support does not block inline execution. Same-task agents share one workspace, and the coordinator remains the only Git mutation owner.

The Process

Step 1: Load and Review Plan

1. Read plan file
2. If the plan or active checkpoint includes an Execution Readiness View, read it before implementation and compare the plan against its intent lock, scope fence, baseline lock, owner / contract constraints, compatibility boundary, retirement boundary, test obligations, review gates, drift / rewind rules, and evidence required before completion.
3. Review critically - identify any questions or concerns about the plan
4. If the view contradicts the plan, baseline, or current worktree evidence, return to plan review or refresh the advisory handoff before editing.
5. Run the TDD Route Guard before implementation: confirm Mode, Decision, Strict authority, strict signals, light eligibility, Test posture, and verification. Strict steps require either an explicit user/project request or a recorded auto decision; plan approval or risk labels alone are not authority. An off-mode missing record may be repaired only as Mode: off / Decision: skipped without loading TDD. Missing/unsupported auto decisions return to plan review. An auto-light record is unsupported when any strict signal is present or when its tiny/low-risk/single-owner/no-behavior-change proof is incomplete. Only Decision: strict with recorded strict authority may authorize steps named Write failing test, Verify RED, GREEN, or REFACTOR. Do not infer strict during execution.
6. If concerns: Raise them with your human partner before starting
7. Before the first write, capture TaskStartSnapshot: root, HEAD, branch or detached state, upstream divergence, staged/unstaged/untracked paths, active Git operations, and git worktree list --porcelain. Preserve task-preexisting state; do not stash, reset, clean, or commit it.
8. Reuse the current branch unless rules require independent history or another goal owns it. If justified, switch/create it in the current workspace when safe; a worktree still requires concurrent checkout or blocking dirty state.
9. If no concerns: Create TodoWrite and proceed

Step 1.5: Long-Task Checkpoint Setup

If the plan has multiple tasks, may span sessions, or includes architecture / contract / workflow changes:

1. Announce: "I'm using the long-task-continuation skill to keep this plan checkpointed and drift-aware."
2. Load aegis:long-task-continuation.
3. Create the initial checkpoint from the plan:
   • current todo
   • active task
   • completed tasks
   • evidence refs
   • blockers
   • next step
4. Before each task, restate the current checkpoint.
5. After each task, update checkpoint, evidence refs, and drift check.

Before a verification-driven unplanned edit, read retained PatchShape, CanonicalOwner, UpwardDrillSignal, outcome, and evidence refs. Route their comparison with the candidate to systematic-debugging before editing; it decides whether the directions converge. A proven independent canonical-owner root stays on the normal plan path.

Step 2: Execute Tasks

For interface/interaction changes, compose ui-ux-governance before affected edits and verify the plan's scoped experience criteria at the relevant user seam.

For each task:

1. Mark as in_progress

2. Follow each step exactly (plan has bite-sized steps)

3. Before any new source-code path is added by a task, restate the plan's Change Necessity or create a compact one if the plan failed to carry it forward. Plan approval is not by itself proof that a new helper, small guard, new branch, fallback, adapter, or owner is necessary.

   Change Necessity:
   - User-visible need:
   - No-change / non-code option:
   - Why code change is necessary:
   - Minimum change boundary:
   - Decision: no-change | docs/config-only | code-change | needs-clarification

   If the decision is not code-change, pause execution and return to plan review instead of editing. If the decision is code-change, carry the minimum boundary into the edit and verification scope.

4. Before any non-trivial source edit, run the plan's Pre-Edit Complexity Check or create a compact one:

   Use using-aegis/references/complexity-governance.md for shared artifact classes, pressure signals, and over-budget handling.

   Complexity Budget:
   - Artifact class:
   - Target files / artifacts:
   - Current pressure:
   - Projected post-change pressure:
   - Budget result: within-budget | at-risk | over-budget
   - Planned governance:
   
   Pre-Edit Complexity Check:
   - Target edit file:
   - Existing pressure signal:
   - Safer edit boundary:
   - Decision: edit-in-place | extract helper | add owner file | split task | pause for plan update
   
   Pre-Edit Owner-Fit Decision:
   - Edit intent: wiring-only | move-out / extract-first | local-fix-without-new-responsibility | new-responsibility | emergency / compatibility patch
   - Owner fit:
   - Safer edit boundary:
   - Decision: edit-in-place | extract helper | add owner file | split task | pause for plan update

   If the check contradicts the plan's file boundary, pause and return to plan review instead of silently stuffing logic into an overloaded owner. If the budget result is over-budget and the task does not also govern that overrun, stop execution and return to plan review rather than pushing the task through as if it were still atomic. When the target edit file is over-budget or mixed-purpose, new-responsibility must not be added in place by default. wiring-only, move-out / extract-first, and local-fix-without-new-responsibility may proceed only when they do not add a new responsibility and the verification boundary is clear. emergency / compatibility patch requires residual risk and a retirement trigger.

5. Run verifications as specified

6. The coordinator is the Git mutation owner. After the coherent Task passes its planned verification, use verification-before-completion before a default local commit, stage only task-owned paths, and read back HEAD, the committed file list, and remaining task delta. no commit, read-only, no-change, and failed-verification tasks create no normal commit.

7. Update TodoCheckpointDraft and DriftCheckDraft before marking the task completed. When an Execution Readiness View exists, the drift check must explicitly compare the active slice against the view's intent lock, scope fence, baseline lock, compatibility boundary, retirement boundary, test obligations, and review gates.

8. Mark as completed