learn
Skill1 starsChanged 3 months agoMITView on GitHub
Store a correction the user made on a draft as a standing rule, so it is honoured next session instead of re-corrected. Use when they push back on how something is written.
194 other projects use a version of this.
Adopted by
Most recently taken up by
- blauwtje/exotheir own version, first seen Nov 16, 2026
- Robo-Co-op/robobuilder-standardtheir own version, first seen Nov 16, 2026
- BloodSweatxED/Claude-skillstheir own version, first seen Nov 16, 2026
- kristenschumann/claude-skillstheir own version, first seen Nov 16, 2026
- Atmoskun/claude-skillstheir own version, first seen Nov 16, 2026
- gfunc/skillstheir own version, first seen Nov 16, 2026
Used with
Projects with it also use add-anydoc, add-clidash and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
TomGranotSkilladd-anydoc
Add local office-document-to-Markdown conversion to NanoClaw agent containers with the pinned Firecrawl AnyDoc CLI. Use when agents need to read attached Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV, or text-based PDF files without uploading them to a hosted parser.
Also in 5 projects that use this
TomGranotSkilladd-clidash
Add clidash — a zero-dependency, read-only web dashboard that derives its tabs and tables at runtime from any CLI that lists resources as JSON. Ships pre-wired for NanoClaw's ncl CLI (agent groups, sessions, channels, users, roles), plus message-activity charts, a log tail, and a read-only file viewer for group skills/CLAUDE.md/profiles.
Also in 5 projects that use this
seek888Skillopen-gstack-browser
Launch GStack Browser — AI-controlled Chromium with the sidebar extension baked in. Opens a visible browser window where you can watch every action in real time. The sidebar shows a live activity feed and chat. Anti-bot stealth built in. Use when asked to "open gstack browser", "launch browser", "connect chrome", "open chrome", "real browser", "launch chrome", "side panel", or "control my browser". Voice triggers (speech-to-text aliases): "show me the browser".
Also in 11 projects that use this
Saqlain-811Skillios-sync
Regenerate the iOS debug bridge against the latest upstream gstack templates. (gstack)
Also in 6 projects that use this
MicrosoftSkilladd-codex
Use Codex (CLI + AppServer) as the full agent provider — planning, tool orchestration, native compaction, MCP tools, session resume — in place of the Claude Agent SDK. ChatGPT subscription or OPENAI_API_KEY. Per-group via agent_provider. Distinct from using OpenAI as an MCP tool (where Claude remains the planner).
Also in 6 projects that use this
MicrosoftSkilladd-deltachat
Add DeltaChat channel integration via @deltachat/stdio-rpc-server. Native adapter — no Chat SDK bridge. Email-based messaging with end-to-end encryption.
Also in 6 projects that use this
- +5 more from TomGranot
- +1 more from microsoft
Other versions194 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 194
- garrytan/gstack★ 142ksame name
- rohitg00/ai-engineering-from-scratch★ 79ksame name
- gavrielc/nanoclaw★ 31ksame name
- nanocoai/nanoclaw★ 31ksame name
- qwibitai/nanoclaw★ 31ksame name
- tigerless-labs/autoharness★ 11ksame name
- tw93/claude-health★ 7.6ksame name
- tw93/Waza★ 7.6ksame name
- nykooi1/vibe-wise★ 3.5ksame name
- HughYau/AcademicForge★ 2.6ksame name
184 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository19 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- company-researchSkill
- cover-letterSkill
- cv-tailorSkill
- interview-coachSkill
- job-scoutSkill
- negotiatorSkill
- network-mapperSkill
- openspec-apply-changeSkill
- openspec-archive-changeSkill
- openspec-exploreSkill
- openspec-proposeSkill
- openspec-sync-specsSkill
- openspec-update-changeSkill
- outreach-drafterSkill
- setupSkill
- submitSkill
- trackerSkill
- upskillSkill
The file
We split this skill into the rules it gives an agent
12 rules. 0 are common, 7 are distinctive, and 5 are very specific to this project.
0
Common
Most of the 167 edited versions we read keep these.
7
Distinctive
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 24
Do not wait to be asked. But do not run it on every objection either: one round-trip question is the whole gate, and it costs a sentence.
- Distinctive: line 31
Ask it plainly, in those terms.
- Very specific: line 44
Is it theirs, or is it everyone's?: If the rule would be true for any user of Lucy, it belongs in the shipped house spec …
- Very specific: line 50
Is a new rule even the right fix?: Read docs/FAILURE-MODES.md.
- Distinctive: line 57
What class does the correction belong to?: If the user objects to a specific word, ask what it is an instance of, and write the rule at …
- Very specific: line 82
If there is no honest mechanical form, store it as prose.
- Distinctive: line 100
Never a paragraph, and never a promise that it will "remember" — point at the file.
- Distinctive: line 118
Only checking them means the model writes the wrong thing and then patches it.
- Distinctive: line 126
Retiring a rule is a normal thing to do: If a stored rule keeps being overridden in practice, say so and offer to delete it.
- Very specific: line 128
If a prose rule has been corrected twice more since it was stored, that is the signal to find its mechanical form and add a check: to it.
- Distinctive: line 135
Never write a rule into a company folder or into a single document.
- Very specific: line 137
Never edit career_facts.yaml from here.
“Common” is measured: kept by at least 6 in 10 of the 167 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
learn — turn a correction into a rule that survives the session
Why this exists
A correction that only fixes the draft in front of you dies with the session. The same correction then arrives again the following week, and the user experiences that as the agent forgetting. It is not forgetting. The rule was never written anywhere it would be read again.
Nobody can state their writing rules up front. They discover them by rejecting drafts. So the
profile has to be able to grow after setup, and this skill is how it grows.
When to run it
Whenever the user pushes back on generated writing — a CV bullet, a letter beat, an outreach
message — and the pushback is about how they want things written rather than about a fact
being wrong. A wrong fact is a career_facts.yaml problem, not a rule.
Do not wait to be asked. But do not run it on every objection either: one round-trip question is the whole gate, and it costs a sentence.
The one question that decides everything
Is this a one-off for this document, or should I keep to it from now on?
Ask it plainly, in those terms. Then:
- One-off → fix the draft, store nothing, say nothing more about it. Most corrections are one-offs and a store full of one-offs is worse than no store.
- Standing → fix the draft and store the rule, in this session, before moving on.
Never infer. Silently promoting every objection to a permanent rule accumulates contradictions that nobody can later explain, and the user ends up fighting rules they never knowingly agreed to. Never quietly drop one either: if they said it stands, it gets written down before the session ends.
Before you write it down: three checks
- Is it theirs, or is it everyone's? If the rule would be true for any user of Lucy, it
belongs in the shipped house spec (
profile/06-cover-letter-notes.mdabove the YOURS line, or the CV spec), not in one person's store. Say so, and offer to raise it upstream. The store is for taste, standing refusals and personal register; it is not a place to park universal craft rules where only one user benefits. - Is a new rule even the right fix? Read
docs/FAILURE-MODES.md. Most repeat corrections are structural, not knowledge gaps: a spec re-derived instead of linked, a missing template, a review that arrived too late, the drafter grading its own work. For those, a new rule is the fix that works least often, and adding one makes the spec longer without making the output better. Name the class in the entry'sclass:field. A healthy spec gets shorter over time. - What class does the correction belong to? If the user objects to a specific word, ask what it is an instance of, and write the rule at that level. "No em dashes" returns as an en dash. "No dash used as punctuation" does not. Then pin the legitimate neighbours with a test so the widening does not overshoot: banning "passionate" must not catch "compassion".
Writing it
python scripts/learned_rules.py add \
--id no-spelled-out-units \
--scope cv,letter \
--rule "Write quantities with numerals and symbols. 30%, not thirty percent." \
--why "Corrected on the Revolut draft, 21 Aug." \
--class instance-scoped \
--pattern '\b(per ?cent|percent|dollars)\b' \
--message "spelled-out unit; use % or \$"--scopeiscv,letter,outreach,all, or a comma-separated mix. Scope it as narrowly as the correction actually justifies; a rule that fires on documents it was never meant for is how a store becomes noise the user learns to ignore.--whyis close to required in practice. A rule whose reason nobody remembers is a rule nobody can safely retire, so it stays forever and the spec only grows.--patternis optional and worth real effort. A rule in prose is followed most of the time; a rule that exits non-zero is followed every time. If a mechanical form exists, add it.- If there is no honest mechanical form, store it as prose. It still reaches the drafter
through
brief. Do not invent a fragile regex just to have one — a check that fires on legitimate text gets the whole store ignored.
Then prove it works before telling the user it is stored: run
python scripts/learned_rules.py check <the draft they just corrected> --scope <scope> and
confirm it fires on the text they objected to. A check nobody has watched fail is not known to
work. Two checks in this repo's cover-letter checker shipped silently inert for exactly that
reason.
Telling them
One line. What was learned, where it lives, and that they can delete it.
Stored that: quantities as numerals, for CVs and letters. It is in
profile/learned-rules.yamland it is now checked, so a draft that breaks it fails rather than reaching you. Delete the line if you change your mind.
Never a paragraph, and never a promise that it will "remember" — point at the file. The learning is legible or it is not trustworthy.
The other half: reading the rules back
Storing is useless if nothing reads them. Every drafting skill runs this before writing:
python scripts/learned_rules.py brief --scope letterand after writing, as part of its checker run:
python scripts/learned_rules.py check <draft> --scope letterBoth halves are load-bearing. Only reading them means they are followed most of the time. Only checking them means the model writes the wrong thing and then patches it. Doing both is what makes a learned rule behave like a shipped one.
Maintenance
python scripts/learned_rules.py list— everything stored, with dates and whether it is checked or prose.python scripts/learned_rules.py validate— the store is well formed.- Retiring a rule is a normal thing to do. If a stored rule keeps being overridden in practice, say so and offer to delete it. Ask before deleting; it is the user's file.
- If a prose rule has been corrected twice more since it was stored, that is the signal to find
its mechanical form and add a
check:to it.
Hard rules
- The store is the user's file. It is gitignored, it never leaves their machine, and nothing goes into it that they have not agreed to in words.
- Never write a rule into a company folder or into a single document. The test is would this still be true for another company? If yes, it goes in the store.
- Never edit
career_facts.yamlfrom here. Facts and rules are different things, and the honesty gate depends on the facts file changing only when the user's real history does.
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.

