vision-drift-check
Skill11 starsChanged 50 days agoMITView on GitHub
Weekly DorkOS scheduled skill: look for retired phrasing and stale positioning across the site, READMEs, docs, app copy and operating skills, check meta/ROADMAP.md against what shipped, and report what drifted from meta/VOICE.md. Reports only; never edits, commits or opens PRs.
- Installs packages
In this repository44 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .claude/CLAUDE.mdCLAUDE.md
- auditing-uiSkill
- browser-testingSkill
- capturing-product-mediaSkill
- ci-local-exportSkill
- ci-pulseSkill
- ci-quarantineSkill
- ci-statusSkill
- creating-pull-requestsSkill
- debugging-systematicallySkill
- debugging-test-failuresSkill
- debugging-typescript-errorsSkill
- deps-weeklySkill
- designing-frontendSkill
- opensrcSkill
- organizing-fsd-architectureSkill
- runtime-watchSkill
- stewarding-ci-pipelineSkill
- syncing-agent-skillsSkill
- verification-before-completionSkill
- visual-companionSkill
- writing-adrsSkill
- writing-app-copySkill
- writing-blog-postsSkill
- writing-changelogsSkill
- writing-developer-guidesSkill
- writing-for-humansSkill
- writing-to-usersSkill
- adding-config-fieldsSkill
- clarifying-requirementsSkill
- maintaining-dependenciesSkill
- maintaining-dev-playgroundSkill
- managing-specsSkill
- marketplace-devSkill
- orchestrating-parallel-workSkill
- reading-session-transcriptsSkill
- receiving-code-reviewSkill
- requesting-code-reviewSkill
- styling-with-tailwind-shadcnSkill
- test-driven-developmentSkill
- upgrading-runtime-dependenciesSkill
- working-in-worktreesSkill
- dorkbot-triageSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 20
Run the two hard gates and note whether they pass:
- Distinctive: line 27
Run the candidate scan:
- Very specific: line 35
Check
meta/stays clean: only the north-star set and current canon at the top level, every file listed in meta/INDEX.md, no new file that …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
vision-drift-check
The hard gates (scripts/check-banned-words.sh, scripts/check-vocab-gate.ts) fail the build on words that are wrong everywhere. This job catches the softer drift they cannot: "open source" in a headline, "local first" on the site, a roadmap item described as working, a north-star file that fell behind what shipped. It runs every Monday at 14:00 UTC and only reports.
The source of truth for every judgment is the north-star set: meta/VOICE.md (words and claims), meta/ROADMAP.md (what is built and what may be claimed), meta/VISION.md and meta/PRINCIPLES.md.
Run
- Work in a throwaway copy of
main, never the shared checkout:git fetch origin, pinBASE=$(git rev-parse origin/main), clear any copy a killed run left behind (git worktree remove --force "$TMPDIR/vision-drift" 2>/dev/null; git worktree prune), thengit worktree add --detach "$TMPDIR/vision-drift" "$BASE"and run every step below from there. Remove it withgit worktree remove "$TMPDIR/vision-drift"when the report is written. The gates neednode_modules; runpnpm install --frozen-lockfile --prefer-offline --ignore-scriptsin the copy first. Run the two hard gates and note whether they pass:
bash scripts/check-banned-words.sh pnpm check:vocab-gateRun the candidate scan:
bash scripts/check-vision-drift.sh- Judge each candidate line against
meta/VOICE.mdandmeta/ROADMAP.md. Most hits are fine in context (a license section may say "MIT license"; a comparison page may describe a rival's "side by side" view). Keep only real drift. - Check the roadmap against reality. For each ticket the scan lists, read its state in Linear (team DOR, read-only, through the flow
linear-adapterskill, orcomposio execute LINEAR_* --account dorkos: thetrackerAccountin.agents/flow/config.local.json, neverartblocks). Flag a ticket that is Done whilemeta/ROADMAP.mdstill lists it as before launch or roadmap, and a "Built today" claim whose feature was removed. - Check
meta/stays clean: only the north-star set and current canon at the top level, every file listed inmeta/INDEX.md, no new file that restates the word lists or the demo-claim gate instead of linking to them.
Report
End the run with one short report, in plain words:
- Gates: pass or fail, with the first failing line if any.
- Drift found: each real problem as
path:line, what it says, and whatmeta/VOICE.mdormeta/ROADMAP.mdsays instead. Group by surface (site, README and npm, docs, app, operating skills, meta). - Roadmap out of date: tickets whose state disagrees with
meta/ROADMAP.md. - Nothing found is a fine report. Say so in one line.
If there is real drift, send the owner one short note with the counts and the top three items through the DorkOS tool whose name ends in relay_notify_user. Do not post in rooms, comment on tickets, edit files, commit or open pull requests: a person or a builder session decides what to change.
Approving it
DorkOS finds the schedule: block above and lists this job on the Schedules page as Waiting for approval. It does not run until the operator approves it. It runs Bash and reads Linear, so approve it with Approve at Full autonomy; with plain Approve every run ends Blocked (docs/guides/task-scheduler.mdx).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

