feature-radar（特征雷达）

Feature Discovery & Prioritization

Mode Routing

Route rules:

• full: Follow normal workflow (Phase 1-6).
• quick: Execute Phase 1-3, then skip to Completion Summary.
• evaluate: Run "Subsequent Runs" reconciliation (steps 1-2), then jump directly to Phase 5-6.
• focus (#N): Read .feature-radar/opportunities/{N}-*.md. If not found, list available opportunities and ask user to pick. Then run Phase 5 for that single opportunity, followed by Phase 6.

State the detected mode before proceeding: "Mode: {mode}"

Bootstrap (First Run)

If .feature-radar/ does not exist at the project root, create it:

.feature-radar/
├── base.md
├── archive/
├── opportunities/
├── specs/
└── references/

Then generate base.md by completing the following:

1. Detect stack — read go.mod, package.json, Cargo.toml, pyproject.toml. Follow imports to understand the dependency graph. Don't just read the config — understand the architecture.
2. Map structure — list top-level directories. For each, read at least one file to understand its purpose. Identify: entry points, core logic, tests, docs, configuration.
3. Extract features — scan exports, commands, API routes, or public functions. Read the implementations, not just the names. Understand what each feature actually does.
4. Find inspiration sources — read README, CONTRIBUTING, docs/ for related projects and communities
5. Verify with user — present the generated base.md and ask: "Does this accurately describe your project?"

After presenting base.md, support iterative refinement per references/WORKFLOW-PATTERNS.md.

base.md Template

Generate base.md following the structure defined in references/SPEC.md (sections 2-5). Read references/SPEC.md first, then fill {placeholders} with project-specific analysis.

Key sections to generate:

• Project Context — detected language, architecture, key feature areas, core philosophy, inspiration sources
• Feature Inventory — implemented features table + value & innovation landscape table
• Tracking Summary — counts per category (start at 0)
• Directory Layout — tree view
• Classification Rules — per references/SPEC.md §3.2-3.5
• Maintenance Workflow — feature flow between directories
• Archive Extraction Checklist — per references/SPEC.md §4.3

After creating the directory, ask the user: "Should I add .feature-radar/ to .gitignore?" (Recommended if the tracking data is internal and shouldn't be committed.)

Subsequent Runs

On subsequent runs (.feature-radar/ already exists):

1. Read existing base.md — do NOT overwrite
2. Run reconciliation per references/DEEP-READ.md steps 2-6
3. Proceed to Mode Routing

Behavioral Directives

Additional directives for this skill:

• Do not skip phases outside mode rules — follow the Mode Routing and Phase execution HARD-GATEs. For conditional phases, state the skip condition check result before deciding to skip.
• Reconcile on subsequent runs — see "Subsequent Runs" section above.

Workflow

Execute phases in order.

full mode (default):

• Phase 1-3: ALWAYS execute.
• Phase 4: Skip ONLY if no documentation directory exists.
• Phase 5-6: Skip ONLY if no open opportunities exist.

quick mode:

• Phase 1-3: Execute. Stop after Phase 3 → Completion Summary.

evaluate mode:

• Phase 1-4: Skip (reconciliation already done in Subsequent Runs).
• Phase 5-6: Execute. Skip ONLY if no open opportunities exist.

focus mode (#N):

• Phase 1-4: Skip.
• Phase 5: Evaluate the single targeted opportunity only.
• Phase 6: Propose for that opportunity only.

For each phase completed, state what was produced before moving to the next phase.

Phase 1: Scan & Classify

1. Read source files (feature ideas, user feedback, ecosystem observations, creative proposals, issue trackers)
2. Cross-reference each entry against the codebase
3. Classify:

- Done / Covered / Rejected / N/A → archive - Open / Partially Done → opportunity - Cross-cutting pattern → specs - External observation → references

Checkpoint: Present classification results using this format, then ask "Continue to Phase 2?"

Phase 1 complete:
| Classification | Count | Items |
|---------------|-------|-------|
| Archive       | {n}   | {list} |
| Opportunity   | {n}   | {list} |
| Spec          | {n}   | {list} |
| Reference     | {n}   | {list} |

Phase 2: Archive Completed Features

For each archive candidate:

1. Create archive/{nn}-{slug}.md
2. Run the Archive Extraction Checklist (mandatory):

- Extract learnings → specs/ - Derive new opportunities → opportunities/ - Update references → references/ - Update ecosystem trends → specs/ecosystem-trends.md

1. Mark the entry as processed in the source (strikethrough + status)

Phase 3: Organize Open Opportunities

1. Create opportunities/{nn}-{slug}.md
2. Assess Impact and Effort realistically
3. Write an honest "Our Position" — do we actually want this?

Checkpoint: Present opportunities using this format, then ask "Continue to Phase 4?"

Phase 3 complete: {n} opportunities organized
| # | Opportunity | Impact | Effort | Our Position |
|---|------------|--------|--------|-------------|
| {nn} | {title} | High/Med/Low | High/Med/Low | {1-line stance} |

Phase 4: Gap Analysis

Find implemented features that docs don't mention:

1. Scan documentation for coverage of implemented features
2. For each gap: which feature, which page should cover it, standalone guide or section

Phase 5: Evaluate & Prioritize

Rank into tiers:

• Build next: High value + strong demand or innovation potential + reasonable effort
• Build soon: Good value + moderate demand
• Monitor: Low demand or premature
• Skip: Conflicts with philosophy or negligible value

Checkpoint: Present tier ranking using this format, then ask "Continue to Phase 6 (Propose)?"

Phase 5 complete:
| Tier | # | Opportunity | Demand | Value | Innovation | Effort/Impact | Fit | Timing |
|------|---|------------|--------|-------|------------|--------------|-----|--------|
| Build next | {nn} | {title} | {H/M/L} | {H/M/L} | {H/M/L} | {H/M/L} | {H/M/L} | {H/M/L} |
| Build soon | ... | | | | | | |
| Monitor    | ... | | | | | | |
| Skip       | ... | | | | | | |

Phase 6: Propose & Decide

For each "Build next" feature (top 1-3), present a proposal card:

### {nn}. {Title}
**Pitch**: {One paragraph — what value does this create?}
**Effort**: {N days} — {brief justification}
**Key decisions**:
- {decision 1}
- {decision 2}

After all cards, ask: "Should we enter plan mode for [feature]?"

Completion Summary

Follow the template in references/DIRECTIVES.md, with skill name "Complete" and an additional line: Top recommendation: {feature name} — {one-line pitch}

Guardrails

• Don't copy blindly. Evaluate fit with YOUR architecture and users, not someone else's.
• Don't overcount. 1 issue with no comments = weak signal.
• Don't undercount. Multiple independent asks = strong signal.
• Chase value, not features. Ask "what problem does this solve?" before "what does this do?"
• Be honest about effort. Low < 1 day. Medium 1-3 days. High 1+ week.
• Challenge deferred items. "Deferred" ≠ "forever" — re-evaluate each session.
• Think creatively. The best features aren't always the obvious ones — look for novel angles.