explorable
Skill0 starsChanged 15 days agoApache-2.0View on GitHub
Use when the person wants to understand something or have it explained to others, where trying teaches more than reading, asks for an interactive page, or wants to try options with their own numbers. A decision page accompanies the decision brief, never replaces it.
4 other projects use a version of this.
Adopted by
Most recently taken up by
- geekplux/explorabletheir own version, first seen Nov 1, 2026
- MbappeWU/explorabletheir own version, first seen Sep 26, 2026
- gmchande/explorable-skillstheir own version, first seen May 8, 2026
- EvanOman/claude-skills-publictheir own version, first seen Mar 12, 2026
Other versions4 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 4
- EvanOman/claude-skills-public★ 4same name
- gmchande/explorable-skills★ 4same name
- MbappeWU/explorable★ 2same name
- geekplux/explorable★ 0same 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 repository43 other files it gives its agents.
- AGENTS.mdAGENTS.md
- decision-briefSkill
- diagnoseSkill
- explain-specSkill
- gather-requirementsSkill
- hand-off-ticketsSkill
- land-changeSkill
- review-councilSkill
- review-findingsSkill
- run-acceptanceSkill
- slice-ticketsSkill
- sync-trackerSkill
- tests-worth-keepingSkill
- using-outcomeboundSkill
- decision-briefSkill
- diagnoseSkill
- explain-specSkill
- explorableSkill
- gather-requirementsSkill
- hand-off-ticketsSkill
- land-changeSkill
- review-councilSkill
- review-findingsSkill
- run-acceptanceSkill
- slice-ticketsSkill
- sync-trackerSkill
- tests-worth-keepingSkill
- using-outcomeboundSkill
- adopt-outcomeboundSkill
- decision-briefSkill
- diagnoseSkill
- explain-specSkill
- explorableSkill
- gather-requirementsSkill
- hand-off-ticketsSkill
- land-changeSkill
- review-councilSkill
- review-findingsSkill
- run-acceptanceSkill
- slice-ticketsSkill
- sync-trackerSkill
- tests-worth-keepingSkill
- using-outcomeboundSkill
The file
We split this skill into the rules it gives an agent
6 rules. 0 are common, 3 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 71
Interview: Ask the questions the work waits on, drawn from the open forks (gather-requirements) or from a spec's rows (explain-spec): one …
- Distinctive: line 85
Keep the controls to the values that move the answer.
- Distinctive: line 104
Every node is one idea, and every line carries information the layout does not already show.
- Distinctive: line 121
Ask each question once. A what-if input that restates a brief's choice asks it twice.
- Very specific: line 123
Read an explorable through check.
- Very specific: line 136
Write the content and the page code over the starter.
“Common” is measured: kept by at least 6 in 10 of the 3 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
Explorable
An explorable is one HTML file that the person opens from disk: they change values, step through
flows and read charts. It is of one of four kinds. You write its source; the engine joins it to a
shared shell and checks it. A page that teaches comes first: learning teaches the person in this
session, explain teaches readers who are not in it, decision helps the person decide, and
interview asks the questions the work waits on. You are done when the person can open the page
and use it: outcomebound explorable check reports no FAIL (its shell, hosts and parts read
PASS; its run line is always UNVERIFIED); the page's own check, run in a browser you drive,
shows no script error and every expectation passing, its result is copied into the working file
beside the page, and your report names the browser tool and what the result held (with no browser
to drive, the run is UNVERIFIED and your report says so); every number the page's lesson or answer rests on
has an expectation; your report names each output that reads UNVERIFIED; and the page's path is in the text the person reads, beside the decision brief
when there is one. You have looked at the page, and it shows what the person asked for, with data
that supports that meaning, not only a matching label; where the data cannot, your report says so.
When a page earns its place
Make one only where text serves the person badly:
- something to understand, or to explain to others, where seeing it move or trying it teaches more than reading;
- a decision whose answer turns on values the person may judge differently from you (rates, sizes, dates, weights, risk), or on how the options work in ways text compares poorly;
- a set of questions that is easier to answer on a page than in the chat;
- the person asks for one.
A request to understand or to explain something gets a learning or explain page, never a decision page. Do not turn an explanation into a request to choose, a recommendation, or a quiz unless the person asked for it. A report, survey or comparison with controls is an explain page, whatever word the person used; it holds a recommendation only where the person asked for one.
A choice the brief settles alone gets no page. A decision page never replaces the brief: draw the
brief as the decision-brief skill says and put it in the text, then name the page beside it. A
page holds only the work that waits on the person; the rest goes on while they read it.
What each kind holds
Learning. Teaches the person in this session. Start from what they say they know and let them
skip it. Start with a concrete case, then state the rule it shows. Use numbers where quantities
matter, a model where changing inputs helps, and step-through where order matters. Show each result
beside the input that caused it. Invite them to try something (set the rate to zero and watch)
before you explain what happens. Questions are optional: add a prediction before a reveal, or a
recall question later, only where the person wants to test themselves. A page with a question has the reply; one with none has no reply. Report what they
answered, never that they understand (explain-spec holds the same rule).
Explain. Teaches readers who are not in this session, who cannot ask you and may open the page later. Teach as a learning page does, and also: say at the top who it is for and what they can do after reading; define each term where it first appears; assume nothing from the session; give each value its source. It has no reply, no brief and no recommendation unless the person asked for one, and it needs no question. Before you call it shareable, read it for private facts (names, paths, customer data, credentials, internal ids, unreleased plans), remove or generalise them, and say what you removed. Sharing the file is the person's act.
Decision. In this order: the briefs (build draws them from the brief document);
each option's mechanism, drawn so that what differs between options shows first; the measures the
choice turns on, each with its value for each option and where the value came from; what-if inputs
for the values the person may judge differently or that are unknown; for each input that decides
the choice, how far it must move before the answer changes (explorable.threshold); a timeline
where time or reversibility differs between options, with the point after which an option cannot
be undone; a question that asks the person to imagine the leading option has failed and say why;
and the reply. Leave out everything that does not change the decision, however true: background,
history and parts no option touches.
Interview. Ask the questions the work waits on, drawn from the open forks (gather-requirements)
or from a spec's rows (explain-spec): one idea in each, the most important first, options of equal
length, your recommended answer marked and never preselected. The runtime adds "I don't know" to
every question; keep it, since such an answer marks a gap.
Guards
A page can mislead as well as inform. Hold these:
- Label every value you supply with its source: checked, assumed, or the person's.
- An unknown value stays empty, and every result that needs it says it is unknown. Never fill a gap with a plausible number.
- An uncertain value is a range, two inputs for the low and the high, never one exact number.
- Show only the precision the inputs support.
- Keep the controls to the values that move the answer; the threshold shows which those are.
- On a decision page your recommendation is already in the brief, so the page shows what would change it, not more reasons for it.
- Motion starts only when the person starts it.
- Colour never carries meaning alone: a label, mark or shape goes with it.
- Page code never navigates, opens a window, fetches, or loads anything; the page's data is in the page. The policy stops loads and sends but not navigation, so this one is yours to hold.
- What the person enters stays in the browser, where Chrome lets any page opened from disk read it. Where the answers are private, tell the person to press "Forget what I entered" when they finish.
Diagrams
- Draw a diagram only when the reader must see order, flow, dependency, state or change; a table says the same about attributes, and a sentence about one relation.
- For options, draw one flow per option with the same names for the same parts, so the difference stands out; give each kind of work (on a schedule, on a change, on a request) its own class and a legend. Where an option changes today's system, draw it against today, marking what it adds, removes and changes.
- Every node is one idea, and every line carries information the layout does not already show.
- Accent one or two elements, the ones the decision turns on.
- A diagram that needs a guide to read becomes an overview and a detail.
- Give each diagram a title and a one-sentence description for screen readers.
Write it once
Rework and repeated text take the person's time and add nothing to the page:
- Outline the page's parts and interactions, then write the content straight into the source. Drafting the content in full in your reasoning first writes it twice; working out a number or a choice there is not drafting.
- Take content from the most condensed document the work already has, with each value's source as that document gives it, and open a longer source only for the part a section needs.
- Where many tables or cards share one shape, keep their rows as data in the page's script and
build them in page code, so that a change edits one row. A citation stays a link in the source:
checkrefuses an address in a script. - Ask each question once. A what-if input that restates a brief's choice asks it twice; give what-if inputs only to values that no brief chooses.
- Read an explorable through
check. Where layout matters, such as a diagram's width against the column, measure it in a browser you drive; with none, the layout isUNVERIFIED. - Change the source with targeted edits; never write it out again to change one part.
Build and check
- Work in the project's working area (
.agents/work/<task>/where the workspace fragment applies), never in a design folder or a commit unless the person asks: a page can hold private facts. outcomebound explorable new <name>.source.html --kind decision(orlearning,explain,interview) writes a starter source; for a decision it also writesbriefs.jsonbeside it when that file is missing. For a decision page only, the brief document is the oneoutcomebound briefreads: fill it, and draw it withoutcomebound brief briefs.jsonfor your message.- Write the content and the page code over the starter.
references/runtime.mdnames theexplorableobject, the attributes and the reply's form. - For each computed output the lesson or answer relies on, add
explorable.expectwith input values and the output each must show. Work the expected value out by hand or with another tool, never by running the page's code, and keep that working in a file beside the source. outcomebound explorable build <name>.source.htmlwrites<name>.html. When it refuses the source, fix the source and build again; never edit the built page, which the next build replaces.outcomebound explorable check <name>.htmlchecks the shell, the hosts and the parts. Its run line is alwaysUNVERIFIED: the engine starts no browser. Run the page yourself in a browser tool the harness gives you (an in-app browser, Playwright, a browser extension): open the address that line names, which ends in#explorable-check, wait until the element with the idexplorable-check-resultshows no library as"pending", and read it asreferences/runtime.mdsays. Where the tool shows a file from disk without running its scripts, serve the page's folder on the loopback address (python3 -m http.server --bind 127.0.0.1), open the page there with the same fragment, and stop the server after. Copy the result's text into the working file beside the page. A script error or a failed expectation is a defect to fix in the source. Do not start a browser from the shell for this: a browser started and killed that way can leave its processes and files behind.- Name the page's path in your message, beside the brief where there is one. Open the page for
the person with the system's opener (
openon macOS,xdg-openon Linux,wslviewunder WSL; on WindowsInvoke-Itemorstart "", which have not been run) only when they ask, or when they are at the machine and the work waits on them.
The reply
A decision, an interview and a learning page with questions end in a reply; an explain page and a
learning page with no question have none. The person pastes a block that references/runtime.md describes. Read each choice line as their
answer to that brief, by its id. Read each input line as their value: where it changes the
answer, work the answer out again with it and say what changed. Record each answer where the work
is recorded; an "I don't know" is a gap to settle, not a blank. A reply grants only what the chosen
option names; an act a permission check guards still needs the person's own words for it.
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.

