sync-horizon（同步水平线）

Sync Horizon

Synchronize eve-skillpacks with the latest state of the eve-horizon repository.

Core Insight

System docs often lag behind implementation. Features ship via feat commits with plans marked "Implemented" — but docs/system/ may not update for weeks. This sync goes to the source: plans, code changes, and commit messages tell us what actually shipped. System docs are one signal among many, not the primary driver.

Prerequisites

• The eve-horizon repo must be at ../eve-horizon (sibling directory)
• .sync-state.json must exist in the repo root (create from template if missing)
• .sync-map.json must exist in the repo root

Architecture: Orchestrator + Parallel Workers

You (the orchestrator) discover what shipped and dispatch focused workers. Each worker handles one update in isolation with its own context budget.

The orchestrator reads lightweight signals (commit log, plan headers, file stats, CLI module list). Workers read the heavy content (plan bodies, code diffs, source docs, target files).

Output Standards

• Distill only shipped platform behavior. Plans marked Implemented/Complete describe shipped features.
• Do not carry roadmap content into skillpacks. Ignore plans marked Proposed/Draft and sections like Planned (Not Implemented), What's next, or "current vs planned" framing.
• Keep eve-work/eve-read-eve-docs/SKILL.md task-first: route by intent, load only the minimal reference files needed.
• Keep reference docs scoped and actionable — curated distillations, not copies.

Workflow

Phase 1: Deep Discovery (orchestrator)

Read lightweight signals from multiple dimensions to understand what actually shipped.

1a. Commit log

cd ../eve-horizon && git log --oneline <last_synced_commit>..HEAD

Categorize commits:

• feat: — new capabilities (primary interest)
• fix: — bugfixes that may affect documented behavior
• docs: — documentation changes (check if they describe already-shipped features)
• chore: / refactor: — usually no skillpack impact, but note removals

1b. Plan intelligence

This is the richest signal. Plans describe what shipped, why, and how.

cd ../eve-horizon && git diff --stat <last_synced_commit>..HEAD -- docs/plans/

For each changed plan, read its header only (first 15-20 lines) to extract status:

cd ../eve-horizon && head -20 docs/plans/<plan-file>.md

Categorize:

• Shipped: Status contains Implemented, Complete, or Done — these are the primary source of truth for new capabilities
• In progress: Status contains In Progress, Partially Implemented — note for awareness but don't distill unfinished work
• Proposed/Draft: Status contains Proposed, Plan, Draft, Ready to build — skip entirely
• Removed: Plan files deleted in this range — something was deprecated or abandoned

IMPORTANT: Do not read full plan bodies. Just headers. Workers will read the plans they need.

1c. Code signal

New CLI commands, DB migrations, and package changes confirm what shipped and reveal capabilities that plans may not fully describe.

# New/changed/removed CLI commands
cd ../eve-horizon && git diff --stat <last_synced_commit>..HEAD -- packages/cli/src/commands/

# New DB migrations (table/column names reveal capability shape)
cd ../eve-horizon && git diff --stat <last_synced_commit>..HEAD -- packages/db/migrations/

# Key package changes (new modules, removed modules)
cd ../eve-horizon && git diff --stat <last_synced_commit>..HEAD -- packages/shared/src/ packages/api/src/ packages/worker/src/ packages/gateway/src/ --stat-name-width=120 | head -50

1d. System doc changes

Still a useful signal — especially for removals and corrections.

cd ../eve-horizon && git diff --stat <last_synced_commit>..HEAD -- docs/system/ AGENTS.md

1e. Current CLI surface

Snapshot the current CLI command modules. This is the ground truth for what agents can invoke.

cd ../eve-horizon && ls packages/cli/src/commands/*.ts

Phase 2: Capability Synthesis (orchestrator)

STOP and think. This is the critical step. Cross-reference all signals to build a capability map.

For each shipped plan (Implemented/Complete), create a capability entry:

Capability: <name>
  Plan: docs/plans/<file>.md (status: Implemented)
  Feat commits: <matching commit hashes>
  Code signal: <new CLI commands, new migrations, new/removed modules>
  System doc: <updated/created/unchanged>
  Scope: <which platform areas this touches>
  One-line summary: <what this capability does, from the plan header>

Also create entries for capabilities evidenced by feat commits that have no corresponding plan — these are smaller features or fixes that shipped directly.

Also note removals: plans deleted, CLI commands removed, modules deleted. These need to be cleaned from our skills too.

Phase 3: Cascade Analysis (orchestrator)

For each capability in the map, determine its impact on eve-read-eve-docs:

3a. Reference doc impact

For each capability, ask:

1. Does an existing reference already cover this area? → Worker updates it
2. Is this a new primitive that needs its own reference? → Worker creates it
3. Was something removed? → Worker cleans stale content from affected references

Cross-reference against the current reference index in eve-work/eve-read-eve-docs/SKILL.md and the reference_docs section of .sync-map.json.

3b. SKILL.md routing impact

Check if eve-work/eve-read-eve-docs/SKILL.md needs:

• New entries in the Task Router
• New entries in the Index
• New rows in the Intent Coverage Matrix
• New trigger keywords in the frontmatter
• Removal of stale entries for deprecated capabilities

3c. Other skill impact

Check .sync-map.json skill_triggers and composite_triggers for other skills affected. Also consider:

• eve-se/eve-manifest-authoring — if manifest shape changed
• eve-se/eve-deploy-debugging — if deploy/infra behavior changed
• eve-se/eve-auth-and-secrets — if auth model changed
• eve-work/eve-agent-memory — if storage primitives changed
• eve-design/eve-fullstack-app-design — if app patterns changed

Phase 4: Plan Work Items (orchestrator)

For each update identified in Phase 3, create a work item:

• Title: Update <target-file>: <capability name>
• Description — everything a worker needs to operate independently:

- The eve-horizon repo path (../eve-horizon) - The commit range: <last_synced_commit>..HEAD - Which plan(s) to read (the shipped plans relevant to this update) - Which source files to diff or read (code, system docs) - The target file path to update or create - Whether this is a reference doc update, skill update, new reference, or removal - The appropriate worker instructions from below

Add a final work item: Update sync state and produce report — blocked until all updates finish.

If any work item touches eve-work/eve-read-eve-docs, also add:

• Run state-today compliance scan for eve-read-eve-docs
• Validate progressive-access routing in eve-read-eve-docs/SKILL.md
• Update.sync-map.json with any new reference doc mappings

Phase 5: Dispatch Workers (parallel)

Spawn one background worker per work item. Launch them all at once.

Each worker prompt must be self-contained. The worker has no access to the orchestrator's conversation. Include:

1. The plan file(s) to read — these are the primary source of truth
2. The git diff command for relevant code/doc changes
3. The target file to read and modify
4. The capability summary from Phase 2
5. The appropriate worker instructions from below

Worker Instructions: Reference Doc Update

Your primary sources are shipped plans (marked Implemented/Complete) and code changes. System docs are secondary — they may lag behind or be absent. 1. Read the plan(s) listed in your task — understand what shipped and why. 2. Read the current reference doc you're updating. 3. Run the git diff for relevant source files (code, system docs). 4. If the plan references specific CLI commands or APIs, verify they exist: cd../eve-horizon && ls packages/cli/src/commands/<name>.ts 5. Distill the shipped capability into the reference doc: - What it does (from plan + code) - How to use it (CLI commands, manifest config, API calls) - Key constraints or requirements 6. Exclude roadmap content. Only document what is implemented. 7. Edit the existing file; do not rewrite from scratch. 8. Preserve existing structure, voice, and formatting.

Worker Instructions: New Reference Doc

1. Read the plan(s) that describe this capability. 2. Check for any system doc coverage: cd../eve-horizon && cat docs/system/<name>.md (may not exist). 3. Read the CLI command source if relevant: cd../eve-horizon && cat packages/cli/src/commands/<name>.ts 4. Create a reference doc that follows the conventions of existing references in eve-work/eve-read-eve-docs/references/. 5. Include: purpose, CLI commands/flags, manifest config if relevant, key behaviors, constraints. 6. Keep it concise and agent-actionable. These are distillations, not documentation copies. 7. State-today only — no planned/roadmap content.

Worker Instructions: Skill Update

Your primary sources are shipped plans and code changes. 1. Read the plan(s) listed in your task. 2. Read the current SKILL.md you're updating. 3. Update with new commands, changed workflows, or new capabilities. 4. Remove stale content that references deprecated features. 5. Keep state-today only. Ensure progressive disclosure. 6. Maintain imperative voice and conciseness. 7. Edit the existing file; do not rewrite from scratch.

Worker Instructions: SKILL.md Routing Update

This worker updates the eve-read-eve-docs SKILL.md routing, index, and coverage matrix. 1. Read the current eve-work/eve-read-eve-docs/SKILL.md. 2. For each new reference doc created by other workers, add: - A Task Router entry describing when to load it - An Index entry with the file path and one-line description - An Intent Coverage Matrix row if it serves a distinct user intent 3. For each removed capability, clean up stale routing entries. 4. Add any new trigger keywords to the YAML frontmatter. 5. Maintain alphabetical or logical ordering within sections.

Worker Instructions: Staleness Check

This is a cross-cutting skill that spans multiple platform primitives. Your job is to validate that the skill's claims match current platform reality. 1. Read the SKILL.md and all files in its references/ directory. 2. Read the shipped plan(s) listed in your task for context on what changed. 3. List current CLI command modules: cd../eve-horizon && ls packages/cli/src/commands/*.ts 4. Scan the skill for stale claims: - "No dedicated X" / "X not available" / "not yet supported" — check if X now exists. - "Current Gaps" or "Workarounds" sections — verify each gap is still a gap. - Decision tables or comparison matrices — check for missing rows. - References to removed features (e.g., inference/ollama/managed-models). 5. If stale claims are found, update the skill. Add new primitives. Remove false gaps. 6. Keep state-today only. Maintain imperative voice and conciseness. 7. Edit existing files; do not rewrite from scratch. Report what was stale and what you corrected.

Worker Instructions: Sync Map Update

Update .sync-map.json to reflect new mappings discovered during this sync. 1. Read the current .sync-map.json. 2. For each new reference doc created, add a reference_docs entry mapping the source files to the target. 3. For each new skill trigger relationship discovered, add a skill_triggers entry. 4. For removed source docs, clean up stale mappings. 5. Ensure watch_paths covers any new directories that should be monitored.

Example Worker Prompt

You are updating a reference doc in the eve-skillpacks repository.

## Your Task
Update the file: eve-work/eve-read-eve-docs/references/agents-teams.md
Add coverage for agent aliases (short names for chat addressing).

## Capability Summary
Agent aliases let users declare short names for agents instead of using full
project-slug-agent-name. Declared in manifest under `agents[].alias`. Resolved
at chat dispatch time. Status: Implemented.

## Primary Source
Read the plan: cd ../eve-horizon && cat docs/plans/agent-aliases-plan.md

## Code Confirmation
The agents CLI was updated: cd ../eve-horizon && git diff <commit>..HEAD -- packages/cli/src/commands/agents.ts
New migration: packages/db/migrations/00076_add_agent_alias.sql

## Target
Read and edit: eve-work/eve-read-eve-docs/references/agents-teams.md

## Rules
- Distill only shipped behavior from the plan and code
- Edit the existing file; do not rewrite from scratch
- These are curated distillations for agents, not copies
- Keep it concise and actionable

Phase 6: Collect Results and Finalize (orchestrator)

Wait for all workers to complete.

Once all update work items are done:

1. Get current HEAD: cd../eve-horizon && git rev-parse HEAD
2. Run state-today and progressive-access checks: ./private-skills/sync-horizon/scripts/check-state-today.sh
3. Update .sync-state.json:

- Set last_synced_commit to the HEAD hash - Set last_synced_at to current ISO timestamp - Append to sync_log (keep last 10 entries)

Phase 7: Report (orchestrator)

## Sync Report: <old_commit_short>..<new_commit_short>

### Commits
- <count> commits synced (<feat count> features, <fix count> fixes)

### Shipped Capabilities (from plans)
- <capability>: <one-line summary> (plan status, feat commits)

### Capability Cascade
For each shipped capability:
- <capability> → updated <reference/skill>, because <reason>

### Removals
- <what was removed/deprecated and cleaned from skills>

### Updated Reference Docs
- <file>: <what changed and why>

### New Reference Docs
- <file>: <what it covers>

### Updated Skills
- <skill>: <what changed>

### SKILL.md Routing Changes
- <new routes, index entries, or coverage matrix rows>

### Sync Map Changes
- <new mappings added>

### State-Today Compliance
- <pass/fail + scan results>

### Coverage Gaps
- <shipped capabilities with no reference doc yet>
- <plan files with no sync-map entry>
- <CLI commands with no documentation coverage>

### Next Steps
- <manual follow-up needed>

Key Constraints

• Plans are the primary source of truth for understanding what shipped. System docs are secondary. Code changes are confirmation.
• Orchestrator reads lightweight signals only: commit log, plan headers (first 15-20 lines), file stats, CLI module list. Never full plans, full diffs, or source docs.
• Workers read the heavy content: full plans, code diffs, source docs, target files.
• Worker independence: Each worker prompt must be fully self-contained with plan paths, code paths, target path, capability summary, and update rules.
• Parallelism: All workers launch simultaneously. No worker depends on another worker's output (except the routing update worker, which should launch after reference doc workers finish).
• Edit, don't rewrite: Workers modify existing files incrementally.
• State-today fidelity: Only document shipped, implemented behavior.
• Progressive access: Preserve task-first routing in entry skills and keep deep detail in references.