| name | feature-brief |
| description | Create a minimal execution-facing brief for one concrete feature or slice after implementation-planning has identified the current stage or current focus. Anchor the work in discovery.md, system-design.md, and system-map.md, define the slice briefly, record what must be protected, and leave the work ready for TDD-oriented execution. Requires discovery.md, system-design.md, system-map.md, and docs/implementation-plan.md to exist.
|
Feature Brief
Output contract: The skill's conversational output must be in Traditional Chinese. Any newly written brief content produced by this skill must also be in Traditional Chinese unless a downstream file explicitly requires another language.
This skill exists to answer one question:
For the current implementation focus, what exactly are we building now, what matters most to protect, and what is the minimum shared brief the team needs before starting TDD-oriented work?
This skill is not another round of implementation planning.
It takes the current project-level plan and turns one concrete unit of work into a short execution brief.
Not every feature brief has to be pre-listed explicitly in docs/implementation-plan.md.
This skill may also be used for current-stage implementation work that emerges during development, such as:
- an implementation correction
- a seam cleanup
- a refactor slice
- a bugfix slice
- a local enablement task
Use this skill directly when the work still fits inside the current stage assumptions.
If the work changes stage order, stage exit criteria, blockers, or enablement-vs-behavior sequencing, route back to implementation-planning.
Purpose
Use this skill to turn:
docs/implementation-plan.md
discovery.md
system-design.md
system-map.md
into a short, execution-facing brief for one feature or one implementation slice.
This skill is responsible for:
- restating the current work in plain language
- anchoring it in the upstream documents
- defining the slice boundary
- identifying the main risk to protect
- recording the minimum known / unknown execution notes
- recording the validation intent in a compact form
This skill is not responsible for:
- creating a new project-level implementation order
- rewriting upstream documents
- creating a full acceptance-spec document by default
- expanding into a long clarification dossier
Shared Document
Create or update:
docs/features/<feature-slug>/brief.md
This document is the short execution brief for the current work item.
Required Outputs
Before finishing, you must produce:
Workflow
Phase 1: Read The Current Focus
Read the current implementation focus from docs/implementation-plan.md.
Extract at minimum:
- current stage
- current focus statement
- touched responsibility units
- touched core data / entities
- touched seams
- current stage exit criteria
If the implementation plan does not yet make the current focus clear enough, route back to implementation-planning.
If the current work is not explicitly named in the implementation plan but still clearly belongs inside the current stage, you may proceed.
In that case, note briefly in the feature brief how this work relates to the current stage.
Phase 2: Anchor Upstream
Anchor the work to:
- relevant discovery goal(s) and interaction(s)
- relevant system-design decisions or trade-offs
- relevant responsibility units, core data / key states, entities, interactions, and seams from
system-map.md
Keep this short.
The brief is not another map or another design document.
Phase 3: Define The Slice
Write the current work item in a compact execution form.
Capture at least:
Feature summary
Slice type: enablement / behavior
Start
End
In scope
Out of scope
Main risk
This should be shorter and more concrete than the project-level implementation plan.
Phase 4: Record Execution Notes
Record only what the current work still needs for honest execution.
Use three short buckets:
Known
Unknown
Needs to be pinned down now
If too many unknowns remain, route back upstream instead of pretending the brief is ready.
Phase 5: Record Validation Intent
Capture the minimum shared testing intent for this work.
At minimum include:
Main risk to protect
Scenario bullets
Suggested primary layer: unit / integration / feature / manual
Use short scenario bullets by default.
Prefer one concise bullet per scenario.
If you use Given / When / Then, keep the whole scenario in the same bullet line instead of spreading it across multiple loose lines.
Example: - Given review has dirty segments, when the operator changes page, then the system autosaves before refreshing only the segment list and pagination.
Do not turn this into a full Gherkin document.
Feature Brief Shape
Write or update docs/features/<feature-slug>/brief.md using this minimum shape:
# Feature Brief — [Feature Name]
## Summary
- Feature summary: ...
- Slice type: enablement / behavior
## Upstream Anchor
- Discovery goals / interactions: ...
- Relevant system-design decisions: ...
- Relevant responsibility units: ...
- Relevant core data / entities: ...
- Relevant entity interaction path: ...
- Relevant seams: ...
- Implementation-plan stage: ...
## Slice Definition
- Start: ...
- End: ...
- In scope: ...
- Out of scope: ...
- Main risk: ...
## Execution Notes
- Known: ...
- Unknown: ...
- Needs to be pinned down now: ...
## Validation Intent
- Main risk to protect: ...
- Scenario bullets:
- Given ..., when ..., then ...
- Given ..., when ..., then ...
- Suggested primary layer: unit / integration / feature / manual
Validation
Brief check
Can a developer read this brief and know what to build now without reopening all upstream documents?
Scope check
Is the slice boundary narrow enough to execute honestly?
Risk check
Is the main risk concrete enough that later tests can target it directly?
Escalation check
If important unknowns remain, is it clear whether they belong in:
implementation-planning
system-design
system-map
If not, the brief is still too vague.
Stage-impact check
Does this brief stay inside the current stage, or did it expose a stage-level change?
Route back to implementation-planning when the work reveals:
- a new blocker that changes sequencing
- a missing prior enablement stage
- changed current-stage exit criteria
- a need to reorder later work
Guardrails
- Do not rewrite the implementation plan.
- Do not turn this into a full spec packet.
- Do not restate all upstream reasoning in long form.
- Do not create a separate Gherkin file by default.
- Do not keep unknowns hidden; record them briefly or route back upstream.