agent-native-reviewer
Review agent-native parity for skills, prompts, tools, commands, generated mirrors, repo workflows, and user-facing actions.
Sync root VISION.md from changed human and agent inputs; use when project taste, doctrine, or maintainer judgment should learn from recent plans, docs, skills, reviews, or repeated misses.
$ npx -y skills add udecode/dotai --skill sync-vision --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/sync-visionContext preview
The summary Claude sees to decide when to auto-load this skill.
Sync root VISION.md from changed human and agent inputs; use when project taste, doctrine, or maintainer judgment should learn from recent plans, docs, skills, reviews, or repeated misses.
name: sync-vision description: Sync root VISION.md from changed human and agent inputs; use when project taste, doctrine, or maintainer judgment should learn from recent plans, docs, skills, reviews, or repeated misses. disable-model-invocation: true
Handle `$ARGUMENTS`.
`VISION.md` is living project taste. Treat it like the project doctrine layer: learn from repeated human answers, agent misses, plans, docs, skills, research, reviews, and handoffs, then ignore noise.
This skill is the incremental sync lane. It does not replace a `vision` router. It keeps root `VISION.md` current by analyzing changed durable inputs since the last synced commit. Root `VISION.md` is the mandatory essential read and single project vision source.
Use `autogoal` before mutable work. This is a derived autogoal workflow.
packs when available.
advancement.
and N/A rows.
completion semantics, output-budget discipline, and `check-complete.mjs`.
doctrine patch rules, status semantics, and baseline advancement.
Goal handle shape:
Sync vision; done when changed inputs are classified, VISION.md is patched or reaffirmed, baseline semantics are recorded, and checks pass; plan docs/plans/<path>.md.
Durable state lives in:
docs/sync/vision/status.json docs/sync/vision/runs/<date>-<base>-to-<target>/
`status.json` fields:
been classified as `captured`, `reaffirmed`, `rejected`, `run-specific`, `owner-routed`, or `deferred-with-question`.
Never advance `lastSyncedCommit` merely because artifacts were written. Advance only after the committed range is fully accounted for. Working-tree overlays are visible in artifacts but are not baselined until committed.
Working-tree overlays include relevant untracked files. New doctrine docs, plans, rule files, and root docs must be visible in artifacts before commit, but they are not baselined until committed.
Projects may customize the helper with:
docs/sync/vision/config.json
Supported fields:
Keep config project-local. Do not put private URLs, account IDs, or temporary branch state in dotai.
working-tree overlay count. No writes.
Do not advance baseline.
high-confidence reusable decisions, route non-vision owners, and advance baseline only if all committed inputs are classified.
proves the range is fully accounted for.
Use the helper for accounting from the project root:
node .agents/skills/sync-vision/scripts/collect-vision-diff.mjs --status node .agents/skills/sync-vision/scripts/collect-vision-diff.mjs --dry-run node .agents/skills/sync-vision/scripts/collect-vision-diff.mjs node .agents/skills/sync-vision/scripts/collect-vision-diff.mjs --advance --plan docs/plans/<plan>.md
The helper writes:
Analyze changed durable inputs, not the whole repo every time:
Generated `.agents/skills/**` mirrors are not primary input when a source rule or top-level `skills/**` source exists. Use generated mirrors only for sync audits.
Every candidate line or theme becomes one of:
package docs, product spec, behavior spec, or migration guide instead of `VISION.md`.
not advance the baseline unless the range can be safely accounted for without it.
Promote to root `VISION.md` when the rule is reusable project doctrine: global taste, source order, cross-boundary law, public API doctrine, proof standards, review attention, owner-specific doctrine, or supervisor stop conditions.
Do not promote:
1. Load `autogoa
Shared skills for coding agents. Skills are the main routing layer. This repo is the canonical source for reusable udecode workflows; downstream repos should link to these skills instead of copying long SKILL.md files around.
Repo: udecode/dotai
Review agent-native parity for skills, prompts, tools, commands, generated mirrors, repo workflows, and user-facing actions.
Create, verify, repair, and close durable Codex goals with measurable outcomes, evidence gates, plan templates, blocker handling, completion audits, and…
Remove a feature completely with no backward compatibility; delete surfaces, callers, tests, docs, fallbacks, stubs, and dead exports.
Run a scoped Linear backlog autonomously as a sequence of maximal safe parallel batches by composing orchestrator, autogoal, and task. Use when the user wants…
Turn the current Codex thread into a coordination thread that routes implementation work to durable reusable child threads in disposable worktrees with…
Resolve GitHub PR review feedback with source-backed triage, fixes, autogoal plan state, focused proof, replies, and thread resolution.