writing-plans

Files (3)

• SKILL.md
• expanded-planning-guidance.md
• plan-document-reviewer-prompt.md

SKILL.md (first lines)

<EXPLICIT-MODE-GATE>
If activation mode is explicit (`~/.config/aegis/config.toml` has
`activation_mode = "explicit"`, or `AEGIS_ACTIVATION_MODE=explicit` is visible
in the environment) and the current user request did not explicitly invoke
Aegis or this skill by name, exit back to the fast path: answer concisely
without this workflow's checklist, ceremony, or document requirements. If the
user explicitly named Aegis or this skill, proceed normally.
</EXPLICIT-MODE-GATE>

# Execute

→ Existing parent plan/spec and a tiny execution slice? → **Use Planless Slice Lane.**
→ Mechanical or bounded change with no durable boundary (no new owner,
  contract, schema, public API, migration, or compat surface)? → **Use
  Planless Slice Lane without a parent document.**
→ Approved spec/requirements for a new workstream or an escalation trigger? →
  **Write an implementation plan for an engineer with no prior task context.**

For a durable plan: confirm scope and acceptance, map current owners/files,
record the TDD route, decompose into executable tasks, define verification and
retirement, self-review, save, then select the execution route. Proceed unless
a real authorization or safety boundary requires the user.

Bind tasks to the requested user outcomes: object, actions, conditions, and
deliverables. Include the final delivery step and evidence at the user-visible
seam, not only implementation/tests. Keep unmet needs and dependencies explicit;
project-wide goals do not authorize work outside this request. Reuse the
existing scope/acceptance slots; no extra goal artifact is required.

Escalate from Planless Slice Lane when the work adds a new owner, contract,
schema, public API, architecture boundary, migration, persistence,
security/permission, distribution/release surface, or an unclear verification
boundary.

# Writing Plans

This skill turns approved requirements into a bounded, executable plan. The
plan must answer: what changes, why code is necessary, which existing owner and
files change, what stays compatible, what verifies the result, what risk
remains, and what old path is retired or deliberately retained. A plan is
method-pack guidance; it cannot grant authoritative completion.

## TDD Route Guard

Before task decomposition, record `TDD Route` with mode (`off | auto`), decision
(`strict | light | skipped`), authority, test posture, reason, and verification.

Strict RED / GREEN steps belong only to an explicit user/project TDD request or
`TDD Route: strict`. In `off`, use `skipped` unless explicit strict authority
overrides it. An approved plan or a risk label alone is not strict authority.

In `auto`, select `strict` for behavior, bugfix, shared/core, contract,
persistence, permission, migration, producer/consumer, or meaningful
regression signals. Select `light` only when the work is tiny, low-risk,
single-owner, has no behavior change or strict signal, and has an obvious
focused check. Absence of an explicit user TDD request is never evidence for
`light`; the auto decision itself must be recorded.

If risk remains unknown, return to requirements, debugging, or plan review
before task decomposition; do not infer a TDD route from uncertainty.

Only a `strict` decision with stated authority may prescribe failing-test,
RED, GREEN, and REFACTOR steps. Otherwise plan the minimum change plus
diagnostic reproduction or post-change regression. A missing auto decision
returns to route selection before implementation tasks.

## Entry And Artifact Choice

**Announce at start:** on the plan-writing route, say that this skill is
creating the implementation plan. On `Planless Slice Lane`, announce the lane instead and do not claim a plan is being created.

**Execution context:** Reuse the current branch/workspace by default. A branch
needs independent history; a worktree needs concurrent checkout, blocking
unrelated dirty state, or explicit user/repository authority.

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

**Default plan path:** `docs/aegis/plans/YYYY-MM-DD-<feature-name>.md`. Plans do
not go in `work/`; user and repository authority override the default.

Exception: use `Planless Slice Lane` when an existing parent plan/spec already
owns the current tiny execution slice, or when the change is mechanical or
bounded and needs no parent document. Do not save a new plan. Emit a compact
`Slice Card`:

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

On the no-parent branch, `Parent plan/spec:` is `none — direct bounded request`.

## Aegis Project Workspace

Workspace creation is lazy and follows project authority. The Aegis Method
Pack repository must not create or ship a live `docs/aegis/` workspace. When a
durable plan must initialize or update another project's workspace, read the
workspace section of `expanded-planning-guidance.md`. It owns the
`<aegis-workspace-helper>` and `INDEX.md` command detail, not the decision to
create a plan.

## Default Plan Surface

Compact output contract: express these as natural plan content, not a stack of
cards: `Aegis Visibility`, approved scope and plan basis, required baseline
refs, files/owners, compatibility boundary, `Change Necessity`, TDD route,
tasks, verification, risks, and retirement. Keep conditional structures silent unless their trigger below fires.

`Aegis Visibility` is normally one sentence explaining which owner, contract,
retirement, compatibility, or verification pressure makes a durable plan
useful. Structured trace belongs only to audit, debug, release, long-task
review, or explicit request.

Before tasks, perform a `Requirement Ready Check`: identify the approved source,

From GanyuanRan/Aegis, MIT.

More from GanyuanRan/Aegis