writing-plans

Execute

→ Have an existing parent plan/spec and a tiny execution slice? → Use Planless Slice Lane.

1. Emit a compact Slice Card: goal, parent plan/spec, files, boundary, verification, stop
2. Update the parent workstream checkpoint/evidence/drift state if persistent state is needed
3. Do not save a new plan for the micro-slice → Have approved spec/requirements for a new workstream or an escalation trigger? → Write implementation plan. Assume engineer has zero context.
4. Scope check: fact/assumption/unknown, baseline, Ripple Signal Triage, compatibility boundary, dual-track needs
5. File map: what files created/modified, clear boundaries, follow existing patterns
6. Bite-sized tasks (2-5 min each): exact file paths, complete code, exact commands, expected output
7. Self-review: spec coverage, placeholders, type consistency, compatibility, verification, dual-track
8. Save → offer execution choice (subagent-driven or inline) → Plan must answer: problem, baseline, files, compat, verification, risks, retirement. → Escalate from Planless Slice Lane to a durable plan when the slice adds a new owner, contract, schema, public API, architecture boundary, migration, persistence, security/permission, distribution/release surface, or unclear verification boundary.

Writing Plans

Overview

Write comprehensive implementation plans assuming the engineer has zero context for our codebase and questionable taste. Document everything they need to know: which files to touch for each task, code, testing, docs they might need to check, how to test it. Give them the whole plan as bite-sized tasks. DRY. YAGNI. TDD. Frequent commits.

Assume they are a skilled developer, but know almost nothing about our toolset or problem domain. Assume they don't know good test design very well.

This skill is the canonical planning workflow for multi-step implementation work. Use it to convert approved specs or requirements into plans that are executable, testable, impact-aware, and bounded by compatibility and authority constraints.

Announce at start: "I'm using the writing-plans skill to create the implementation plan."

Context: This should be run in a dedicated worktree (created by brainstorming skill).

Input: approved requirements, a Spec Brief, or a Design Spec.

Save plans to: docs/aegis/plans/YYYY-MM-DD-<feature-name>.md Plan always goes to plans/ — never to work/. (User preferences for plan location override this default.)

Exception: if an existing parent plan/spec already owns the current tiny execution slice, use Planless Slice Lane. Do not save a new plan. Emit a compact Slice Card in the conversation or the active long-task checkpoint instead:

Slice Card:
- Goal:
- Parent plan/spec:
- Files:
- Boundary:
- Verification:
- Stop:

If docs/aegis/ does not exist and configured Aegis workspace support is available, initialize the target project first:

python <aegis-workspace-helper> init --root <target-project-root>

If installed Aegis workspace support is unavailable, initialize the workspace manually:

1. Create docs/aegis/README.md and docs/aegis/INDEX.md
2. Create docs/aegis/BASELINE-GOVERNANCE.md from template
3. If the project has code, create docs/aegis/baseline/YYYY-MM-DD-initial-baseline.md Then save the plan and append to docs/aegis/INDEX.md. Prefer:

python <aegis-workspace-helper> append-index --root <target-project-root> --path docs/aegis/plans/<filename>.md --kind plan --title "<title>"
python <aegis-workspace-helper> check --root <target-project-root>

Scope Check

If the input is a Spec Brief, keep the plan scoped to the pinned what/why/acceptance and do not expand into a formal design unless new architecture, contract, migration, or cross-module uncertainty appears.

Compact output contract before writing the plan: Plan Basis, BaselineUsageDraft, Files, Compatibility, Architecture Integrity Lens, Plan Pressure Test, Plan-Time Complexity Check, Tasks, Risks, and Retirement. Expand only where the approved scope, risk, or verification surface requires it.

Use a compact BaselineUsageDraft whenever the plan depends on specific baseline docs or current-authority refs:

BaselineUsageDraft:
- Required baseline refs:
- Delivered context refs:
- Acknowledged before plan refs:
- Cited in plan refs:
- Missing refs:
- Decision: continue | needs-baseline-readback | needs-verification | pause-for-user | blocked

Delivered context refs is optional host-projected bookkeeping only. It is not authoritative proof that a host injected or the model internally consumed a context payload. The artifact exists to make baseline/context attention drift visible before and during planning.

Use the Architecture Integrity Lens before task decomposition when an executable plan may still encode responsibility overlap, a wrong canonical owner, a caller-side fallback, a stale path carrying real logic, or a missed higher-level owner / contract / source-of-truth simplification. Keep it compact: invariant, canonical owner / contract, responsibility overlap, higher-level simplification, retirement / falsifier, and verdict.

Use a compact Plan Pressure Test before task decomposition:

Plan Pressure Test:
- Owner / contract / retirement:
- Architecture integrity / higher-level path:
- Verification scope:
- Task executability:
- Pressure result: proceed | revise plan | return to design

The pressure test is not an approval gate and should not redesign an approved spec without cause. It exists to catch owner / contract / retirement risk, missing verification, and tasks that are too vague to execute safely.

Use a compact Plan-Time Complexity Check before writing task steps when the plan changes maintained source files, core owners, handlers, routers, managers, shared utilities, adapters, or fallback paths:

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

Complexity Budget:
- Artifact class:
- Target files / artifacts:
- Current pressure:
- Projected post-change pressure:
- Budget result: within-budget | at-risk | over-budget
- Planned governance:

Plan-Time Complexity Check:
- Target files:
- Existing size / shape signals:
- Owner fit:
- Add-in-place risk:
- Better file boundary:
- Recommendation: edit-in-place | extract helper | add owner file | split task | defer refactor

If the projected budget result is over-budget, do not write an atomic task that silently assumes add-in-place growth. Revise the task boundary, add governance work, or explicitly mark the slice as requiring follow-up before implementation begins.

If the spec covers multiple independent subsystems, suggest breaking into separate plans. Before writing tasks, check: fact/assumption/unknown, baseline docs, compatibility boundary, whether dual-track (repair + retirement) applies. If approved requirements or the design carried an ADR signal, preserve the ADR signal, source refs, real alternatives, compatibility boundary, and expected baseline-sync questions for completion so ADR Auto Backfill can run without rediscovering the decision from scratch.

If task decomposition would encode a new owner, duplicate owner, fallback, adapter, compat-only carrier, delete-first question, unverified assumption, or long-term stability claim that the spec did not already settle, use first-principles-review and its Decision Hygiene Review or Architecture Integrity Lens before task decomposition.

When the plan must decide between deleting old internal paths, retaining compat for a proven external boundary, or stopping for persistent-state confirmation, compose anti-entropy-governance. Keep it as a narrow classification and guardrail owner; it does not authorize destructive execution.

Use Planless Slice Lane before writing or saving a plan when all of these are true:

• a parent spec or parent plan already defines the workstream
• the current request is executing or refining one bounded task from that parent
• no new owner, contract, schema, public API, architecture boundary, migration, persistence, security/permission, distribution/release surface, or unclear verification boundary appears
• the slice can be described by a Slice Card

The lane preserves long-task continuity without turning execution bookkeeping into durable planning artifacts.

Aegis Project Workspace

Workspace creation is triggered by the plan save step. See using-aegis/SKILL.md Rule 3 for the hard binary rule. If the project already has docs/adr/ or architecture docs, reference them — do not duplicate authority.

File Structure

Map files before defining tasks. Design units with clear boundaries and single responsibilities. Files that change together should live together. Follow existing codebase patterns. Each task should produce self-contained, independently reviewable changes.

Required Planning Outputs

Before you leave this workflow, the written plan must make these items answerable:

1. What problem or approved scope this plan is implementing
2. Which baseline docs, ADRs, or requirements shaped the plan
3. Which required baseline refs were explicitly acknowledged before planning and which were actually cited in the plan
4. What files own the change
5. What compatibility boundary must hold
6. Whether the architecture integrity check found a higher-level owner / contract path before task decomposition
7. What plan-time complexity pressure exists and which edit boundary is safer
8. What verification proves each major slice
9. What risks, rollback surface, old owner/fallback handling, ADR signal preservation, and baseline-sync signals remain

Bite-Sized Task Granularity

Each step is one action (2-5 minutes):

• "Write the failing test" - step
• "Run it to make sure it fails" - step
• "Implement the minimal code to make the test pass" - step
• "Run the tests and make sure they pass" - step
• "Commit" - step

Plan Document Header

Every plan MUST start with: Goal, Architecture, Tech Stack, Baseline/Authority Refs, Compatibility Boundary, Verification. See template in this directory.

Task Structure

Each task: Files (create/modify/test paths), Why (user/business value), Impact/Compatibility, Verification (exact commands), then 5 checkbox steps: Write test → Verify RED → Minimal code → Verify GREEN → Commit. Every step must include complete code and exact commands.

For bug fixes, refactors, contract changes, or governance cleanup, add Repair Track (root cause, canonical owner, minimal sufficient stable repair, compat boundary, verification) and Retirement Track (old owner/fallback, active status, keep reason or deletion trigger) inside the relevant task. If Ripple Signal Triage fired, include the affected downstream consumers and expanded verification path in the same task.

No Placeholders

Never write: "TBD", "TODO", "implement later", "fill in details", "Add appropriate error handling", "Write tests for the above" without actual test code, "Similar to Task N" without repeating code. Every step must contain complete, copy-paste-ready content.

Self-Review

Check plan against spec: 1) Spec coverage — can you point to a task for each requirement? 2) Placeholder scan — any TBD/TODO/vague instructions? 3) Type consistency — do signatures match across tasks? 4) Compatibility — invariants, non-goals, stable interfaces marked? 5) Plan-time complexity and minimality — lowest-entropy owner/file boundary that fixes the bug class, not just the smallest textual diff? 6) Architecture integrity — any higher-level owner / contract / source-of-truth simplification skipped? 7) Verification — exact commands? 8) Dual-track, decision hygiene, and ADR/baseline-sync signals preserved where needed?

Fix issues inline. Re-review is not needed — just fix and move on.

Execution Handoff

After saving the plan, offer execution choice:

"Plan complete and saved to docs/aegis/plans/<filename>.md. Two execution options:

1. Subagent-Driven (recommended) - I dispatch a fresh subagent per task, review between tasks, fast iteration

2. Inline Execution - Execute tasks in this session using executing-plans, batch execution with checkpoints

Which approach?"

If Subagent-Driven chosen:

• REQUIRED SUB-SKILL: Use aegis:subagent-driven-development
• Fresh subagent per task + two-stage review

If Inline Execution chosen:

• REQUIRED SUB-SKILL: Use aegis:executing-plans
• Batch execution with checkpoints for review

Planning Boundaries

• A plan can define implementation slices, verification, rollback surface, and retirement expectations
• A plan cannot grant authoritative completion
• A plan should prepare runtime-ready execution, not pretend to be runtime authority