| name | eagle-feature-flow |
| description | Structured development workflow that sequences planning, building, testing, and shipping. Use when the user says: 'feature flow', 'build this feature', 'implement this', 'dev workflow', 'eagle flow', 'start feature', 'build and ship', 'full dev cycle', 'plan and build', 'implement and review'. Also trigger when the user describes a feature to build and expects the full cycle: planning, implementation, testing, review, and commit.
|
Eagle Feature Flow
A structured development workflow that takes a feature from idea to committed code. Sequences nine phases, orchestrating existing eagle-skills components at each gate.
When to use
- User describes a feature to build (not a bug fix or quick change)
- User says "build this", "implement this feature", "full dev cycle"
- Any task that benefits from plan → build → test → review → ship
The Nine Phases
Execute these phases in strict order. Do NOT skip phases. If a phase fails, fix and re-run that phase before advancing.
Phase 1: Plan
Enter plan mode. Analyze the codebase to understand:
- Where the change fits architecturally
- Which files need to be created or modified
- What dependencies exist
- What could break
State your approach in 2-3 sentences. Get user confirmation before proceeding.
Phase 2: Tasks
Break the plan into discrete tasks with acceptance criteria. Use TaskCreate for each task.
Each task must have:
- Subject: What to do (imperative form)
- Description: Acceptance criteria — specific, testable conditions that define "done"
Example:
Subject: Add rate limiting middleware to /api/auth routes
Description:
AC1: Requests exceeding 10/min per IP return 429
AC2: Rate limit headers (X-RateLimit-Remaining) present on every response
AC3: Existing auth tests still pass
Order tasks by dependency. Set blockedBy relationships where needed.
Phase 3: Build
Implement each task in order. For each task:
- Mark it
in_progress
- Write the code
- Verify acceptance criteria are met
- Mark it
completed
Follow the project's conventions. Read before writing. Keep diffs surgical.
Phase 4: Test
Run the project's test suite. The exact command depends on the stack:
npm test / pnpm test / bun test for JS/TS
pytest for Python
cargo test for Rust
go test ./... for Go
- Whatever
test script is in package.json or the project's test runner
If no test suite exists, write tests for the new code before running them.
Gate: All tests must pass. If any fail, fix and re-run until green.
Phase 5: Spectral Review
Run the relevant review specialists on the changed files. Pick agents based on the stack:
- Backend →
eagle-security-audit, eagle-code-quality
- Frontend →
eagle-ux-code-review, eagle-accessibility-review
- Database →
eagle-database-review, eagle-data-integrity
Gate: Address all HIGH and CRITICAL findings. MEDIUM findings are addressed if the fix is straightforward. LOW findings are noted but not blocking.
Phase 6: Fix
Apply fixes from the spectral review. This may involve:
- Security patches
- Error handling improvements
- Code quality fixes
After fixing, return to Phase 4 (test again) to verify fixes didn't break anything.
Phase 7: Anti-Slop
Run the eagle-anti-slop skill on all new and modified code. Look for:
- Over-abstraction (wrappers nobody needs)
- Over-commenting (comments that restate the code)
- Over-logging (log lines that add no diagnostic value)
- Over-typing (type gymnastics that don't add safety)
- AI-typical patterns (defensive checks for impossible states)
Strip slop. Keep the code lean.
Phase 8: Final Test
Run the full test suite one more time. This catches regressions from Phase 6 and Phase 7 changes.
Gate: All tests must pass. No exceptions.
Phase 9: Commit
Create a git commit (or multiple commits if the work is logically separable).
Commit message rules:
- First line: imperative mood, under 72 characters, describes the why not the what
- Body (if needed): bullet points explaining key decisions
- No "fix: fix the thing" — say what was actually wrong
- No "refactor: refactor code" — say what changed and why
- Include
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
If the work is large, split into logical commits (e.g., one for the feature, one for tests, one for config changes). Each commit should be independently valid — don't commit broken intermediate states.
Do NOT push unless the user explicitly asks.
Phase Flow Diagram
Plan → Tasks → Build → Test ──→ Spectral Review → Fix ─┐
↑ │
└───────────── Test Again ←─────┘
│
Anti-Slop
│
Final Test
│
Commit
Failure Handling
- Test failure (Phase 4/8): Fix the failing tests. Re-run. Do not advance until green.
- Spectral review findings (Phase 5): Fix HIGH/CRITICAL. Re-test. Then advance.
- Anti-slop findings (Phase 7): Strip the slop. Re-test. Then advance.
- Build error (Phase 3): Debug, fix, continue. Do not skip to testing with broken code.
What this skill does NOT do
- It does not push to remote (user must explicitly request)
- It does not create PRs (use eagle-spectral-ship for that)
- It does not deploy (use the project's deployment workflow)
- It does not replace human judgment on the plan — Phase 1 requires user confirmation