navigate（navigate 搜索）

Navigate

*Meta — Utility. Artifact awareness and multi-phase orchestration.*

Core Question: "What exists, and how do I track a complex workflow across sessions?"

Note: Skill routing (matching a user's goal to the right skills) is the agent's job — it reads skill descriptions and proposes matches proactively on every response. Navigate does NOT route. It provides artifact state awareness and persistent workflow tracking that the agent can't do implicitly.

Two Modes

Status

When: Argument is status or user asks "what exists", "what's stale", "what do I have"

Scan .agents/ and the canonical top-level folders (brand/, architecture/, research/), then report. Single-pass, no sub-agents.

1. Scan every .md file in: For each, read frontmatter to extract skill:, date:, status:, version:. Missing fields = —. Most meta-skill artifacts (.agents/meta/) are ephemeral analysis outputs — report them but weigh them lightly in recommendations. Exception: meta/review-chain-report.md is consumed by /ship as a review gate, so treat it as a real dependency when ship is in the workflow. Also scan .agents/meta/out-of-scope/ — report count of out-of-scope decisions and flag any whose "Revisit if" conditions may now be met.

- .agents/ (pipeline artifacts — including .agents/meta/) - brand/, architecture/, research/ (canonical records)

1. Report as a table sorted by date (newest first), with a Location column distinguishing .agents/ pipeline outputs from top-level canonical records. Mark STALE if date: > 30 days old.

| Artifact | Location | Skill | Date | Age | Status |
|----------|----------|-------|------|-----|--------|
| product-context.md | research/ | icp-research | 2026-03-15 | 13d | ok |
| system-architecture.md | architecture/ | system-architecture | 2026-03-10 | 18d | ok |
| solution-design.md | .agents/ | solution-design | 2026-02-10 | 46d | STALE |

If none of the scanned folders exist or all are empty, say so.

1. Recommend the one or two skills that unblock the most, using the dependency graph below. Don't dump a flat list of everything missing — trace the graph, find the root blocker.

Orchestrate

When: Argument starts with orchestrate or user asks for a full workflow plan

1. Scan .agents/ + brand/ + architecture/ + research/ for existing artifacts (same as Status step 1)
2. Check .agents/meta/out-of-scope/ — if the goal overlaps with a prior out-of-scope decision, surface it: "This was previously scoped out because [reason]. Revisit condition: [condition]. Proceed anyway?"
3. Classify the goal, match to skills using the skill registry, check dependencies, identify parallel tracks
4. Produce a workflow-plan.md artifact with phases, checkpoints, and progress tracking
5. Return the plan + instructions for Phase 1

Continuation: Run /navigate orchestrate again (no new goal) to validate current phase, update progress, and recommend next phase.

Orchestrate Output Format

---
skill: navigate
version: 1
date: [today]
status: in-progress
goal: "[user's goal]"
---

# Workflow: [Goal Title]

## Phases

### Phase 1: [Name]
- [ ] /[skill] -> [artifact]
**Decisions:**
- Mechanical: [auto-decided, not surfaced]
- Taste: [auto-decided, surfaced at checkpoint]
- User Challenge: [must ask before proceeding]
**Checkpoint:** [validation criteria]

### Phase 2: [Name]
- [ ] /[skill] -> [artifact]
**Decisions:**
- [same format]
**Checkpoint:** [validation criteria]

## Status
Current phase: 1
Next action: Run `/[first-skill] [context]`

Decision Classification

In Orchestrate mode, classify decisions within each phase into three types:

How to classify:

• If the codebase, spec, or prior artifacts answer it → Mechanical
• If you have a defensible recommendation but the user might disagree → Taste
• If it's expensive to reverse, subjective, or has business implications → User Challenge

In the workflow plan, annotate each phase:

### Phase 2: Architecture
- [ ] /system-architecture -> system-architecture.md
**Decisions:**
- Mechanical: tech stack (matches existing codebase)
- Taste: DB schema normalization level — will default to 3NF, surface at checkpoint
- User Challenge: self-hosted vs managed DB — must ask before proceeding

At phase checkpoints, present Taste decisions as a batch: "I made these calls — confirm or override before I continue." User Challenges block the phase until answered.

Checkpoint Validation

Between phases in orchestrate mode:

1. Existence — required artifacts exist
2. Freshness — date field < 30 days
3. Completeness — valid frontmatter present
4. Status — final preferred; draft triggers warning

Phase [N] Checkpoint: PASS / WARNING / BLOCKED

Cross-Project Mode (--cross-project)

When invoked with --cross-project:

1. Scan .agents/ in current directory AND sibling skill repo directories
2. Report unified table with Project column
3. Trace cross-project dependencies
4. Recommend highest-impact action across ALL projects

Dependency Graph (Canonical)

Artifacts are optional save-points, not mandatory pipeline gates. The graph shows what CAN flow between skills, not what MUST exist on disk. Conversation context from the current session is equally valid.

product-context.md <- /icp-research
|-> market-research.md <- /market-research --+
|-> problem-analysis.md <- /problem-analysis -+
|                                             +-> solution-design.md <- /solution-design
|                                             |   |-> targets.md <- /funnel-planner -> experiment-*.md
|                                             |   |-> mkt/imc-plan.md <- /imc-plan -> mkt/content/ -> mkt/*.humanized.md
|                                             |   +-> system-architecture.md <- /system-architecture
|                                             |       +-> tasks.md <- /task-breakdown
|                                             |           +-> (execute) -> meta/review-chain-report.md <- /review-chain
|                                             |               +-> ship-report.md <- /ship
|                                             |                   +-> deploy-verify-report.md <- /deploy-verify
|-> /discover (conversation context or spec.md) --+
|-> design/brand-system.md <- /brand-system
+-> product/flow/*.md + product/flow/index.md <- /user-flow --> system-architecture.md, tasks.md

Priority Table

When recommending next action. Note: if discover ran in the current session, its decisions are in conversation context — no spec.md file is required.

Pre-Built Workflow Templates

Technical Build

Triggers: "build the app", "implement", "code it"

Phase 1: /discover (interactive conversation — optionally saves spec.md)
Phase 2: /system-architecture -> system-architecture.md
Phase 3: /task-breakdown -> tasks.md
Phase 4: (execution, /review-chain after critical tasks)
Phase 5: /code-cleanup + /technical-writer (parallel)
Phase 6: /ship -> ship-report.md
Phase 7: /deploy-verify -> deploy-verify-report.md

Full Product Launch

Triggers: "launch", "new product", "go to market"

Phase 1: /icp-research -> product-context.md
Phase 2: /market-research + /problem-analysis (parallel)
Phase 3: /solution-design -> solution-design.md
Phase 4: /brand-system + /imc-plan + /funnel-planner (parallel)
Phase 5: /user-flow -> product/flow/<flow-name>.md (once per flow)
Phase 6: /discover (interactive conversation)
Phase 7: /system-architecture -> system-architecture.md
Phase 8: /task-breakdown -> tasks.md + execution (/review-chain after critical tasks)
Phase 9: /ship -> ship-report.md
Phase 10: /deploy-verify -> deploy-verify-report.md
Phase 11: /content-create + /copywriting -> mkt/content/
Phase 12: /lp-optimization + /seo (parallel)

Strategy Sprint

Triggers: "strategy", "what should we build", "prioritize"

Phase 1: /icp-research -> product-context.md
Phase 2: /market-research + /problem-analysis (parallel)
Phase 3: /solution-design -> solution-design.md
Phase 4: /funnel-planner -> targets.md
Phase 5: /experiment -> experiment-*.md

Content Campaign

Triggers: "content campaign", "marketing campaign"

Phase 1: /icp-research -> product-context.md
Phase 2: /imc-plan -> mkt/imc-plan.md
Phase 3: /content-create -> mkt/content/
Phase 4: /copywriting -> mkt/content/*.copy.md
Phase 5: /humanize -> mkt/content/*.humanized.md

Landing Page

Triggers: "landing page", "conversion page"

Phase 1: /icp-research -> product-context.md
Phase 2: /brand-system -> design/brand-system.md
Phase 3: /copywriting -> mkt/content/*.copy.md
Phase 4: /lp-optimization -> mkt/lp-optimization.md
Phase 5: /humanize -> mkt/content/*.humanized.md

Architecture Decision

Triggers: "debate the tech stack", "which approach", "compare options"

Phase 1: /discover -> scope the decision (interactive)
Phase 2: /agent-room debate -> meta/agent-room-report.md
Phase 3: /system-architecture -> system-architecture.md

Anti-Patterns

• Dumping a flat list — Trace the graph, find the root blocker. Don't list every missing artifact equally.
• Ignoring artifact state — Don't recommend /icp-research when product-context.md is 3 days old.
• Skipping dependency tracing — Don't recommend /task-breakdown when system-architecture.md doesn't exist.
• Treating templates as rigid — Skip phases where fresh artifacts exist.
• Using navigate for skill routing — The agent proposes skills proactively. Navigate is for artifact status and orchestration only.

Chain Position

Meta skill — scans artifact state and recommends the highest-impact next skill to run. Start here when unsure what to do next.