# Scoping

> Defines a project's delivery boundary: what is in, what is out, and what is sufficient for the goal. Use when a project's goals are set and its scope needs defining or revising, including scope growth during delivery. Skip single-feature requirements, unresolved intent, and speculative additions at the implementation level.

- Skill: `augments-labs/scoping` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx get-skill add augments-labs/scoping`
- Raw SKILL.md: https://api.skillmd.com/api/skills/augments-labs/scoping/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Marketing & Growth
- Author: augments-labs (https://skillmd.com/u/augments-labs)
- Updated: 2026-11-11
- Page: https://skillmd.com/skills/augments-labs/scoping

---


# Scoping

Name the boundary before anyone starts building.

## When to use

- After the goals are approved, whether they were already present or
  `define-goals` produced them, and before a project boundary that is still
  unresolved is used as an input.
- When scope is unclear, or growing, part-way through a project.
- **Skip** for a single feature. Use `clarifying-intent` only when the
  feature's intent or boundary is ambiguous; `writing-specs` owns its detailed
  requirements and non-goals.

## Step 1: Cut the scope

Open `assets/scope-section.md` before starting. Fill a section only with
what this work has; the template says what an empty one carries.

1. Carry the non-negotiables forward first: goal guardrails, existing
   contracts, preserved behavior and data, security, accessibility,
   compatibility, operability, recovery.
2. Never list a preserved invariant or an existing commitment as out of
   scope.
3. For each assumption and each dependency, record its validation action,
   its owner, and its expiry or decision point. A hidden project inside
   "assumes X" is scope, not an assumption.
4. Write the change rules: the observations that abort this cut, the changes
   that reopen approval, and who decides. If the user is the only
   stakeholder (a solo owner), they decide, and no approver matrix is
   recorded.

## Step 2: Write and present

1. Write the `## Scope` section to
   `.sdlc-skills/briefs/{{YYYY-MM-DD}}-{{topic}}.md` or the user-set path,
   keeping the approved sections around it as they are. Once its identity is
   issued, the section is never edited. To revise, write a successor
   section, headed as the template says, that names its predecessor and
   marks each capability and exclusion
   `added / changed / removed / preserved`; add the predecessor's
   `superseded` row only on the successor's approval.
2. Present:

   ```text
   Scope {{path}} — version {{identity (per template)}}
   In: {{n}} capabilities  Out: {{n}}  MVP: {{one line}}  Assumptions: {{n}} owned

   1. Approve this boundary
   2. Request a scope change
   3. Reject the cut
   4. Cancel

   Recommendation: {{thinnest answer that reaches the goal}} — {{one sentence}}.
   ```

   Ask this and end the turn; `clarifying-intent` owns what closes it.
3. **REQUIRED SUB-SKILL:** on option 1, invoke `writing-specs` when detailed
   requirements are the next missing input. Never impose a phase that is
   already complete. When a successor is approved, list every spec and plan
   bound to its predecessor; work under each stops until its owner
   validates it again.

## Gotchas

- If a later rescoping pass reuses a deferred capability's stable ID for a
  different item, an old reference to that ID, such as a linked task or a
  review note, silently points at the wrong decision. An ID retires; it is
  never reissued.
- An assumption recorded with a validation action but no expiry or decision
  point reads as settled forever. Nothing then re-checks it once whatever it
  depended on changes underneath it.

