harness-creator
Skill49 starsChanged 13 days agoApache-2.0View on GitHub
Create or customize a fak-native agent harness from a user's needs. Use when someone asks to make their own harness, agent product, branded/local UI, provider or tool profile, or a "10-minute" / "weekend" harness build. Drives the shipped external generator, preserves generated/user ownership boundaries, rebuilds and selfchecks the result, and routes deeper UI/extension work through public harnesskit contracts.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- TheAatroxking1/harness_creator_skilltheir own version, first seen Oct 11, 2026
- walkinglabs/learn-harness-engineeringtheir own version, first seen Sep 30, 2026
- tjdwls101010/Harness-Creatortheir own version, first seen Sep 17, 2026
- ffy6511/harness-creator-skilltheir own version, first seen Jul 1, 2026
- idoforgod/cys-harness-creatortheir own version, first seen Jun 2, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- walkinglabs/learn-harness-engineering★ 25ksame name
- TheAatroxking1/harness_creator_skill★ 48same name
- ffy6511/harness-creator-skill★ 4same name
- idoforgod/cys-harness-creator★ 4same name
- tjdwls101010/Harness-Creator★ 4same 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 repository151 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- llms-full.txtllms.txt
- llms.txtllms.txt
- agent-readinessSkill
- appeal-scoreSkill
- bottleneck-mapSkill
- ci-repairSkill
- claim-repro-scoreSkill
- clear-out-wipSkill
- commit-cleanSkill
- conflation-scoreSkill
- cross-validateSkill
- curate-clusterSkill
- debt-cleanSkill
- debt-orchestratorSkill
- disambiguation-scoreSkill
- dojo-rsi-scoreSkill
- dos-dispatch-loopSkill
- dos-dispatchSkill
- dos-next-upSkill
- dos-replanSkill
- dos-witness-claimSkill
- field-borrowSkill
- fleet-waveSkill
- git-subagent-syncSkill
- goalSkill
- guard-rsi-scoreSkill
- harness-gardenSkill
- industry-scoreSkill
- issue-orchestratorSkill
- issue-queueSkill
- issue-triageSkill
- lightgap-scoreSkill
- memory-compactSkill
- milestone-scoreSkill
- modularizeSkill
- negframe-scoreSkill
- opencode-waveSkill
- operator-heaviness-scoreSkill
- persona-scoreSkill
- phased-planSkill
- plan-auditSkill
- quality-scoreSkill
- question-loopSkill
- refresh-cachedoc-numbersSkill
- refresh-readmeSkill
- releaseSkill
- resume-watchdog-auditSkill
- run-it-all-nightSkill
- score-2xSkill
- scorecardSkill
- scout-loopSkill
- signal-firstSkill
- skill-lifecycleSkill
- skill-overlapSkill
- skill-scoreSkill
- slop-scoreSkill
- sota-checkSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
10 rules. 0 are common, 7 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
7
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 11
Create a product around pkg/harnesskit.
- Distinctive: line 15
Write this five-line frame before editing:
- Distinctive: line 32
Ask follow-ups only when an answer changes the profile, policy boundary, interface, or witness.
- Distinctive: line 38
Use this when the need fits product identity, profile/config, policy, instructions, or a thin interface projection.
- Very specific: line 47
Read <target>/harness.lock.json before editing.
- Very specific: line 58
If the timed witness is missing or red, say not yet.
- Distinctive: line 62
When the builder knows the job but not the extension seam, inspect concrete bounded packs first:
- Distinctive: line 76
Use this when the user needs a new provider/model adapter, lifecycle middleware, skill pack, durable state, tool/policy adapter, channel, …
- Distinctive: line 91
Run offline selfcheck before any live model/provider test.
- Distinctive: line 107
When asked what people are discussing on X or elsewhere, use dated primary sources or a reproducible export/search.
“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
/harness-creator — user need to rebuilt fak-native harness
Create a product around pkg/harnesskit; do not fork cmd/fak, import internal/, or parse terminal output. The minimum successful run is an external directory that builds and passes offline selfcheck after one user-owned customization.
Intake: decide only what changes the product
Write this five-line frame before editing:
- For: the person who will use the harness.
- Problem: the job or failure they face now.
- Today: the real next-best alternative, including its setup and maintenance cost.
- Better because: the smallest observable improvement this harness should make.
- Witness: the command and artifact that will prove it.
Classify centrality as Core, Enabling, Stewardship, or Peripheral, then check all four project problems:
| Check | Harness question |
| P1 managed context | What repeated setup, history, or memory should the harness own? |
| P2 net-true efficiency | Against which real alternative will time/cost/quality be measured? |
| P3 bounded adaptation | What may skills, models, and tools change, and what remains fail-closed? |
| P4 integrated operations | How will the user run, inspect, rebuild, upgrade, and recover it? |
Ask follow-ups only when an answer changes the profile, policy boundary, interface, or witness. Prefer a runnable default over a questionnaire.
Choose a track
Ten-minute harness
Use this when the need fits product identity, profile/config, policy, instructions, or a thin interface projection. Promise ten minutes only when a timed clean-room witness supports it.
- Pick an external target directory and Go module path.
Run:
fak harness init --dir <target> --module <module>- Read
<target>/harness.lock.jsonbefore editing. Change only files markeduser; the default customization seam is<target>/product/config.go. - Express the smallest user need by changing field values returned by
product.DefaultConfigand, when needed, the body ofproduct.OfflineReply. Preserve theConfig,DefaultConfig, andOfflineReplysignatures consumed by generated runtime code. Do not editgenerated/orcmd/product/main.go. Rebuild and selfcheck from the target:
go build -o <scratch-or-target-bin> ./cmd/product go run ./cmd/product --selfcheck- Capture: elapsed time, generator version, changed user-owned paths, build result, selfcheck semantic events, and the remaining gap to the stated witness.
If the timed witness is missing or red, say not yet; do not turn the slogan into a claim.
Start from a user-need gallery
When the builder knows the job but not the extension seam, inspect concrete bounded packs first:
fak harness gallery list
fak harness gallery show --id coding-workspace
fak harness gallery init --id coding-workspace --dir ./my-packThe built-ins cover readonly support, a local coding workspace, cited research, and incident operations. Preserve each pack's explicit capability exclusions. A gallery blueprint is a decision aid and starter artifact, not a claim that its weekend integration is already built.
Weekend harness
Use this when the user needs a new provider/model adapter, lifecycle middleware, skill pack, durable state, tool/policy adapter, channel, or custom UI. Start with the ten-minute spine, then add exactly one public seam at a time:
- composition and compatibility: #6792 and #6793;
- headless backend and extension loop:
pkg/harnesskitplus #6803; - skill/instruction packs: #6796;
- custom local web UI: headless protocol
fak.harness.run/v1plus #6790; - packaging and reproducible rebuilds: #6806.
A custom UI consumes semantic events and inputs. It must not scrape CLI/TUI prose or depend on provider wire objects. Keep branding/layout outside the kernel so UI iteration does not require rebuilding fak.
Rebuild and upgrade contract
- Pin fak with
--fak-version; record the immutable version inharness.lock.json. - Rerun
fak harness initto refresh recognized generated files. It must preserve user-owned files byte-for-byte. - Rebuild the external product after every contract or config change.
- Run offline selfcheck before any live model/provider test.
- For a shipped binary, record OS/arch, Go version, module pin, build command, and output hash. Never imply that rebuilding the product rebuilds the fak kernel.
- A protocol-major or conformance change requires migration evidence, not a blind pin bump.
Local UI path
For a usable local fak-native interface, keep three boundaries visible:
- Kernel: stock fak execution, policy, context, and lifecycle behavior.
- Protocol:
pkg/harnesskitsemantic events/inputs, cursor, resume, approvals, and redaction. - Product UI: locally served assets, branding, layout, and interaction state.
The first UI witness is: start locally, complete one offline run, render streamed text and tool/approval state from semantic events, reconnect by cursor, and capture browser renders. Until #6790 lands that witness, report the local web UI as not yet; the existing TUI is not evidence for a separately built UI.
Current-interest research before promotion
When asked what people are discussing on X or elsewhere, use dated primary sources or a reproducible export/search. Inventory at least:
- requests for harness ownership versus IDE/plugin customization;
- local-first/private execution;
- model/provider routing and bring-your-own-model;
- skills, memory, tool/MCP, policy/approval, and multi-agent customization;
- custom UI, generative UI, replay/evals, observability, packaging, and upgrades.
For each theme record date, source URL/export, representative statement, engagement only if visible, and the product decision it could change. Separate observed discussion from product inference. Social popularity is discovery evidence, never proof of usability or value; #6809 owns the independent adoption benchmark.
Promotional frame
Use witnessed language:
- Fast path: “Create a working fak-native harness in 10 minutes” only when init → one owned customization → rebuild → selfcheck is timed on a clean machine.
- Deep path: “Create your own harness this weekend” only when the named extension/UI path has a captured end-to-end result and maintenance boundary.
- Always include: audience, included outcome, prerequisites, measured environment/date, and the command to reproduce it.
A safe pre-witness formulation is: “Target: a rebuilt offline harness in ten minutes; custom providers, skills, and UI are the weekend track.”
Completion receipt
Close with scannable bullets:
- verdict first: working / not yet;
For / Problem / Today / Better because / Witness;- external directory and pinned
module@version; - user-owned paths changed;
- exact build and selfcheck evidence;
- UI state: stock TUI or separately built protocol client;
- filed, deduplicated follow-ups for every remaining gap.
References
docs/harness-init.mddocs/harness-protocol.mddocs/notes/HARNESS-CREATOR-SPINE-WITNESS-2026-08-15.mddocs/spine-first-defaults.mddocs/standards/net-true-value.md- epic #6777; local web UI #6790; skill packs #6796; gallery #6808; independent study #6809
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.

