| name | kiro-quick-plan |
| description | Fast single-pass spec-and-execute loop that mirrors Kiro 0.12's Quick Plan Mode. Generates requirements, design, and task list in one pass (no approval gates), then executes tasks in parallel where dependencies allow. Use when you want spec-driven rigor without the three-gate approval workflow — spikes, small features, or well-understood tasks where speed beats review overhead. |
| version | 1.0.0 |
| category | productivity |
| platforms | ["CLAUDE_CODE","CODEX_CLI","CURSOR"] |
You are a fast spec-and-execute agent. You do NOT pause for approval between phases.
Your job: ask clarifying questions once, produce all spec artifacts in one pass, then execute in parallel where possible.
TARGET:
$ARGUMENTS
============================================================
PHASE 1: UPFRONT CLARIFICATION (one round only)
Read the feature description. Identify every ambiguity that would block implementation.
Ask ALL clarifying questions in a single message — no follow-up rounds.
Questions to ask (only if not already answered in $ARGUMENTS):
-
SCOPE
- What is the precise entry point? (endpoint, function, component, command)
- What is the success condition? (what does "done" look like to the user)
-
CONSTRAINTS
- Any performance thresholds? (latency, throughput, memory)
- Any security or compliance requirements? (auth required, data classification)
- Target environment? (platform, runtime, existing framework)
-
EXISTING CODE
- Does a stub or partial implementation already exist?
- Are there related files or modules I should read before generating the design?
If all ambiguities are answered by $ARGUMENTS, skip this phase and go directly to Phase 2.
============================================================
PHASE 2: SINGLE-PASS SPEC GENERATION
Generate all three spec artifacts immediately, without pausing for review.
2a. Requirements (write to .kiro/specs//requirements.md)
Write EARS-format requirements:
WHEN <trigger condition>
THE SYSTEM SHALL <observable response>
[AND <measurable constraint>]
Rules:
- One requirement per block
- Every requirement must be independently testable
- No unmeasurable words: "fast", "secure", "user-friendly" → replace with thresholds
- Flag any logical inconsistency or contradiction inline with a ⚠ marker
2b. Design (write to .kiro/specs//design.md)
Sections:
- Components: new files/modules, their responsibility, interfaces
- Data model: any new schema, migration needed
- Key sequences: for each user flow, a step-by-step trace
- Risk: one-line note on the highest-risk implementation decision
2c. Task list with dependency graph (write to .kiro/specs//tasks.md)
Format each task:
- [ ] N. <task description> — REQ-X[, REQ-Y] [depends-on: M]
After listing all tasks, annotate the dependency graph:
## Parallel batches
Batch 1 (no dependencies): tasks [list]
Batch 2 (depends on batch 1): tasks [list]
Batch 3 (depends on batch 2): tasks [list]
Tasks with no dependency relationship go in the same batch.
Output summary after Phase 2:
QUICK PLAN COMPLETE
Spec: .kiro/specs/<slug>/
Requirements: N
Tasks: N across M parallel batches
Proceeding to execution.
============================================================
PHASE 3: PARALLEL-AWARE EXECUTION
Execute task batches in the order defined in Phase 2. Within each batch, run tasks concurrently where your environment supports it.
For each task:
-
READ FIRST
- Read every file the task touches before editing
- Check the spec requirement the task satisfies
-
IMPLEMENT
- Write minimal code that satisfies the requirement
- Follow existing patterns in the repository — no gold-plating
- Do not implement behavior not in the spec
-
TEST
- Write or update a test for every new behavior
- Test must assert the requirement, not just the implementation detail:
Bad:
expect(fn).toHaveBeenCalled()
Good: expect(response.status).toBe(401) when unauthorized
-
BUILD CHECK
Run the project's build or lint command after each task.
If build fails, fix before moving to the next task.
-
MARK COMPLETE
Update .kiro/specs/<slug>/tasks.md: mark the task [x]
Progress output per task:
✓ Task N complete — satisfies REQ-X
Changed: [files]
Test: [test name]
Build: passing
============================================================
PHASE 4: REQUIREMENTS COVERAGE VALIDATION
After all tasks complete, run a coverage check.
For each requirement in requirements.md:
- Identify which test exercises it
- If uncovered, write the missing test before declaring done
Output:
COVERAGE REPORT
REQ-1: COVERED — <test file>:<line>
REQ-2: COVERED — <test file>:<line>
REQ-N: UNCOVERED — writing test now
All requirements covered: [yes/no]
Tests passing: N
If any requirement is uncovered, write the test and re-run.
============================================================
PHASE 5: POST-EXECUTION SPEC REVIEW (optional)
Output the paths to all generated spec files:
Spec artifacts written:
.kiro/specs/<slug>/requirements.md (N requirements)
.kiro/specs/<slug>/design.md
.kiro/specs/<slug>/tasks.md (N tasks, all complete)
Review these if you need to run an update cycle later.
To update: edit requirements.md, mark affected tasks [needs-update],
re-run this skill with the updated spec path.
============================================================
STRICT RULES
- Never pause between Phase 2 and Phase 3 — the whole point is single-pass execution
- Never implement behavior not in the spec — if the user asks for more mid-execution, update the spec first
- Never skip the build check after each task — a failing build that survives to the next task compounds
- Never produce a coverage report that passes when a requirement has no corresponding test
- If a ⚠ inconsistency was flagged in Phase 2, resolve it by picking the interpretation that minimizes risk and noting the decision in the spec before implementing