suggest
Skill150 starsChanged 8 days agoMITView on GitHub
Draft a wished change to the framework's core as a suggestion, show the user exactly what it would share, and leave the filing to them. Use on /suggest <wish>, "suggest a change to the core", or when a wish needs a core file that a user's copy keeps read-only. Offers the user's own layer first, checks the draft with the payload gate's probes, and never files the issue itself.
8 other projects use a version of this.
Adopted by
Most recently taken up by
- osmlab/name-suggestion-indextheir own version, first seen Nov 16, 2026
- HurricaHjz/SecondSelfword for word, first seen Nov 15, 2026
- intelligentcode-ai/skillstheir own version, first seen Nov 15, 2026
- mkrz/skillstheir own version, first seen Nov 15, 2026
- blinkboxhq/Blinkboxtheir own version, first seen Nov 8, 2026
- belt-sh/skillstheir own version, first seen Sep 27, 2026
Other versions8 projects use a version of this.
Word for word · 1
- HurricaHjz/SecondSelf★ 150100% same
Edited · 0
None found.
Wrote their own version · 7
- osmlab/name-suggestion-index★ 893same name
- brave/brave-search-skills★ 182same name
- sergeeey/Claude-cod-top-2026★ 9same name
- blinkboxhq/Blinkbox★ 4same name
- belt-sh/skills★ 2same name
- mkrz/skills★ 2same name
- intelligentcode-ai/skills★ 1same name
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 repository15 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- adoptSkill
- atticSkill
- deep-lintSkill
- delegateSkill
- export-templateSkill
- gatherSkill
- ingestSkill
- lintSkill
- outputSkill
- projectSkill
- qmd-searchSkill
- querySkill
- reflectSkill
- weekly-auditSkill
The file
We split this skill into the rules it gives an agent
4 rules. 0 are common, 0 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 6 edited versions we read keep these.
0
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 29
Read .claude/core-paths, one git pathspec a line: a file it matches is core.
- Very specific: line 42
If the user takes that offer, do the job there and stop: no suggestion is needed.
- Very specific: line 82
Exit 2,
PROBE FAILED: Stop. Quote the line. The draft has not been checked, so never call it clean, and go no further until the cause is … - Very specific: line 95
Read repository-code from CITATION.cff: that is the repository's address.
“Common” is measured: kept by at least 6 in 10 of the 6 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
suggest — a wished core change, drafted for the user to file
Contract
CLAUDE.md §12 governs: in a user's copy the agent never edits a core file, and a wished change that the user's layer cannot serve leaves as a suggestion. This skill is that route. The user files the suggestion as a public issue after seeing exactly what it shares, so the agent stops at the command.
Triggers
/suggest <wish> · "suggest a change to the core" · a wish that needs a core file in a user's copy.
What an issue shares
A title, what should change, why, the proposed text, and the core check's line: the fields of the
shipped form, .github/ISSUE_TEMPLATE/core-suggestion.yml. A draft never shares a home path, text
from the wiki or CUSTOMISATION.md, or a transcript. A wish that grew out of the user's own notes is
put in general terms.
Steps
1. Say it is core, and offer the user's layer first
Read .claude/core-paths, one git pathspec a line: a file it matches is core. Say that this is a
core file, in one sentence. Then offer to do the job in the user's layer, which often serves better,
since it works today and survives every update:
| The wish | Its home in the user's layer |
| A preference, a role or a style | CUSTOMISATION.md |
| A new workflow | .claude/skills/my-<name>/, listed in .claude/sanctioned.local.txt |
| A helper | .claude/agents/my-<name>.md |
| A hook | .claude/hooks/my-<name>, its line in .claude/settings.local.json |
| A routing choice | .claude/routing.local.json |
| A setting | .claude/settings.local.json |
If the user takes that offer, do the job there and stop: no suggestion is needed. With no
.claude/core-paths this copy has no locked core, so say so and ask whether a suggestion is still
wanted.
2. Write the draft
Take the date from date +%F. Write output/suggestions/<date>-<slug>.md, in the user's layer,
making the folder if it is absent. If that name exists, add -2: a draft is not overwritten. The
headings are the form's labels, so a filed draft reads as the form does:
# <title>
### What should change
<the change, in the framework's own terms>
### Why
<what it would do for any user>
### Proposed text
<the exact new text, or: none yet>
### Core check
<the session's `core:` line, or what `python3 .claude/hooks/core-check.py --report` prints; where no
check runs in this copy, the words: no check ran>3. Run the draft gate
python3 .claude/skills/export-template/publish_guard.py --draft output/suggestions/<file>It runs the payload gate's own probes on this one file: home paths, personal strings, the extra
lines of output/publish-gate-strings.txt, and credential values. No allowlist applies. The default
agent name is no hit; a name the user chose is.
- Exit 0, clean. Quote its two lines and go on.
- Exit 1, hits. Quote each
FINDINGline. Take a home path, a personal string or a credential value out of the draft. A credential finding names the KEY and never the value; if that value was ever public, the user rotates it. A word that is also a folder name, such asDocuments, may be benign, and so may the public repository's own owner: the user judges those, and the draft goes on only on their word. - Exit 2,
PROBE FAILED. Stop. Quote the line. The draft has not been checked, so never call it clean, and go no further until the cause is fixed.
Any edit to the draft sends it back through this step. The gate has limits: a value under 6 bytes, a case-changed copy and an encoded copy are not caught, and a home path under an unusual mount or on a network share is caught only by the user's own names in it. A person reading step 4 stands behind them.
4. Show the whole text, and say an issue is public
Show the draft whole, word for word, never a summary. Say plainly that an issue is public: anyone can read it, and the user may not be able to take it down later. Ask whether anything should change.
5. Give the link and the command, and stop
Read repository-code from CITATION.cff: that is the repository's address. Give the user the
new-issue link, <address>/issues/new?template=core-suggestion.yml, whose fields take the draft's
sections. Where gh auth status exits 0, also give the exact command for them to run:
gh issue create --repo <owner>/<repo> --title "<title>" --body-file output/suggestions/<file>The user files it. The agent never runs that command, submits the form or files by any other route. Asked to, it says the filing is the user's step and gives the command again. Report the draft's path, the gate's result, and that nothing was filed.
A bug, not a wish
The same steps serve, from step 2, with the form bug-report.yml: its labels are the draft's
headings, and its core check field is required.
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.

