fairladyz625/harness-anything/preset-creator
Claude Code and agent skill by fairladyz625/harness-anything
- community
Checking that the SKILL.md file is still in the repository.
Create, review, or update Harness Anything preset-manifest/v3 packages, including profiles, completion gate selection, template selections, output shapes, script entrypoints, capability declarations, package validation, and preset tests. Use when authoring or migrating a Harness Anything preset.
Install
npx vskill@latest install fairladyz625/harness-anything/preset-creatorSKILL.md
Preset Creator
Use the current contract
Treat packages/preset/src/preset.contract.ts as the contract and packages/preset/assets/software-coding/presets/standard-task/preset.json as the minimal bundled example. Presets are declaration-first packages. Keep preset-specific behavior in package assets or scripts; never add a CLI core branch keyed by preset id, title, or action name.
A self-contained package contains:
PRESET.md uses this frontmatter:
Manifest v3
Use schema: "preset-manifest/v3". Required top-level keys are exactly:
schema,id,title,vertical,version,kind, andoutputShapekernelVersionRangeandcapabilityImportsprofilesanddefaultProfile
Optional top-level keys are extends, policyPath, entrypoints, and completion. No other top-level key is accepted. kind is template-content or process-action; outputShape is a required non-empty string.
Each profile requires id, title, completionGates, and templateSelections; it may add capabilityImports, closeoutOverrides, and archiveOnComplete. completionGates is an array of gate ids declared by the vertical or preset, including an explicit empty array when there are no gates. defaultProfile must name one declared profile. closeoutOverrides accepts boolean values for review, consent, fact, and factDisposition; archiveOnComplete is a boolean. These are task-creation choices, not a way to change an existing task's frozen profile.
The vertical owns domain-wide completion.sources, completion.gates, and completion.closeoutDefaults. A preset may declare additional completion sources or gates in its top-level completion object. The runtime merges both declarations and rejects duplicate source ids, duplicate gate ids, invalid references, and profile selections that do not resolve. Keep the shared domain contract in the vertical and use profiles only to select gate ids.
Each template selection is exactly:
localePolicy.prefer is project, preset, or explicit; fallback is zh-CN or en-US.
Top-level capabilityImports items require id, kind, version, and a boolean required. Profile and entrypoint capability items use id, kind, and version, with an optional boolean required. Valid kinds are checker, scaffold, projection, command, template, and raw-fs.
Minimal valid package
Use this manifest shape:
Validate the package directory, not an individual manifest file:
Continue only when the report has "valid": true and an empty issues array.
Select a declared completion gate
A profile selects gate ids; it does not repeat source schemas or gate contracts:
The selected id must exist in the merged vertical and preset declarations. For a full CSV, PNG, and experiment.json example—including a command entrypoint, a required seed, artifact binding, task submit, rejection, and repair—follow gate-customization. Completion gates govern task completion; use the repository's separate delivery pre-review gate when preparing a submission for review.
Script entrypoints
A v3 entrypoint is a named object with exactly these fields:
Input types are string, number, boolean, or json. intent is a plain string. requires, produces, and sideEffects are flat capability-ref arrays. The command path must name a regular package-local file. Ordinary entrypoints receive their declared context through HARNESS_PRESET_CONTEXT. A command completion source names an entrypoint as <preset-id>/<entrypoint-id>, declares the completion-witness checker capability in both capabilityImports and produces, and reads the frozen witness input through HA_PRESET_INPUT. The script cannot choose its task, execution, source identity, or accepted artifact set.
Migrating v2 packages
Do not relabel a v2 object and leave its fields in place. Apply these mappings:
- Add the required top-level
outputShapewhen absent. - Keep profile ids, titles, string completion-gate ids, and valid current
template selections.
- Convert script
inputsfrom an object/map to the v3 array of
{name,type,required} only when the types and requiredness are explicit.
- Convert a structured/object
intentto one plain string only when its exact
user-visible meaning is present.
- Replace nested capability selectors with flat
requires,produces, and
sideEffects refs only when current providers are explicitly known.
- Discard v2 raw-filesystem
readsandwritespath-glob fields. They have no
v3 manifest counterpart. A raw-fs capability ref names a provider; it does not restore those old path scopes.
- Remove any entrypoint whose authority or input meaning cannot be expressed
without guessing. Preserve its user-visible procedure in PRESET.md and report the removed behavior to the owner.
- Delete all unknown v2 keys. Validation is exact-field and fails closed.
Template assets
Preset manifests select templates; the vertical template catalog owns template metadata and Markdown assets own bodies. Use template://<id>@<version> references. Do not inline template bodies in the manifest. Run ha template list to discover current builtin declarations and validate that every selected slot, path, locale policy, and required anchor resolves.
Workflow
- State the preset job in one sentence and choose
template-contentor
process-action.
- Read the v3 contract and a nearby bundled manifest; do not copy a v2 schema.
- Create
preset.jsonandPRESET.md, then add only declared completion,
policy, template, or script files.
- Run
ha preset validate --source <package-directory> --json. - Run
ha preset install --source <package-directory> --dry-run --jsonbefore
installation and inspect its issues.
- Run
ha vertical validate --source <vertical-directory> --preset <installed-preset-id>
to check the combined completion catalog.
- Add resolver, materialization, permission, or execution tests at the tier
matching the behavior. Put exactly one // harness-test-tier: declaration on the first line of every new Node test file.
Review checklist
- The schema is
preset-manifest/v3;outputShapeis present. - Top-level, profile, selection, capability, input, and entrypoint fields are
exact current-contract fields.
profilesis non-empty anddefaultProfileresolves.- Every selected gate id resolves in the merged vertical/preset declaration;
template selections use the current nested locale policy.
- Command sources point at an installed entrypoint that declares the
completion-witness checker capability.
- No v2
reads,writes, object input/intent, or nested capability selector
survives.
PRESET.mdhas validpreset-document/v1frontmatter.- Package validation reports valid with no issues.
- No preset-specific behavior was added to generic CLI dispatch.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/preset-creator/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/fairladyz625/harness-anything/preset-creator)