Skip to content

DataArt/artisyn-implementation-harness / skills/implementation-harness

implementation-harness

Flagged

Install the portable AI-dev harness, rebuild its operational graph standalone, harvest project skill improvements into its canonical templates, package direct canonical edits into a portable bundle, or apply canonical updates to an existing project. Use /implementation-harness launch, /implementation-harness reindex-graph, or /implementation-harness patch harvest|prepare|apply.

FileKindSize
CHANGELOG.mdmarkdown37.5 KB
README.mdmarkdown13.7 KB
SKILL.mdskill md9.6 KB
bin/artisyn-implementation-harness.mjsexecutablescript11.8 KB
bin/lib/placer.mjsscript2.6 KB
bin/lib/providers.mjsscript1.8 KB
bin/lib/reporter.mjsscript3.0 KB
bin/lib/zip.mjsscript4.0 KB
canonical/adaptors/claude/browser.mdmarkdown2.5 KB
canonical/adaptors/claude/context.mdmarkdown517 B
canonical/adaptors/claude/credentials.mdmarkdown512 B
canonical/adaptors/claude/delegation.mdmarkdown910 B
canonical/adaptors/claude/interaction.mdmarkdown359 B
canonical/adaptors/claude/planning.mdmarkdown745 B
canonical/adaptors/claude/review.mdmarkdown1.0 KB
canonical/adaptors/claude/task-tracking.mdmarkdown897 B
canonical/adaptors/codex/browser.mdmarkdown1.0 KB
canonical/adaptors/codex/context.mdmarkdown517 B
canonical/adaptors/codex/credentials.mdmarkdown480 B
canonical/adaptors/codex/delegation.mdmarkdown910 B
canonical/adaptors/codex/interaction.mdmarkdown359 B
canonical/adaptors/codex/planning.mdmarkdown904 B
canonical/adaptors/codex/review.mdmarkdown973 B
canonical/adaptors/codex/task-tracking.mdmarkdown1.1 KB
canonical/adaptors/copilot/browser.mdmarkdown620 B
canonical/adaptors/copilot/context.mdmarkdown881 B
canonical/adaptors/copilot/credentials.mdmarkdown992 B
canonical/adaptors/copilot/delegation.mdmarkdown1.2 KB
canonical/adaptors/copilot/interaction.mdmarkdown454 B
canonical/adaptors/copilot/planning.mdmarkdown785 B
canonical/adaptors/copilot/review.mdmarkdown791 B
canonical/adaptors/copilot/task-tracking.mdmarkdown723 B
canonical/agents/architect/INTENT.mdmarkdown2.4 KB
canonical/agents/architect/TEMPLATE.mdmarkdown6.5 KB
canonical/agents/docs-writer/INTENT.mdmarkdown2.2 KB
canonical/agents/docs-writer/TEMPLATE.mdmarkdown2.8 KB
canonical/agents/infra/INTENT.mdmarkdown2.1 KB
canonical/agents/infra/TEMPLATE.mdmarkdown5.1 KB
canonical/agents/layer/INTENT.mdmarkdown6.0 KB
canonical/agents/layer/TEMPLATE.mdmarkdown1.5 KB
canonical/agents/onboarding/INTENT.mdmarkdown5.3 KB
canonical/agents/onboarding/TEMPLATE.mdmarkdown4.2 KB
canonical/agents/product-architect/INTENT.mdmarkdown7.5 KB
canonical/agents/product-architect/TEMPLATE.mdmarkdown3.9 KB
canonical/agents/scrum-master/INTENT.mdmarkdown10.9 KB
canonical/agents/scrum-master/TEMPLATE.mdmarkdown12.3 KB
canonical/docs/.og/cli.mjsscript13.6 KB
canonical/docs/.og/cli.test.mjsscript19.6 KB
canonical/docs/.og/constants.mjsscript2.9 KB
canonical/docs/.og/docs-operational-graph.sqlitebinary24.0 KB
canonical/docs/.og/docs.mjsscript5.5 KB
canonical/docs/.og/edges.mjsscript5.8 KB
canonical/docs/.og/schema.mjsscript741 B
canonical/docs/.og/utils.mjsscript5.2 KB
canonical/docs/Engineering Guide/01 Conventions.mdmarkdown3.0 KB
canonical/docs/Engineering Guide/03 Engineer Role.mdmarkdown2.6 KB
canonical/docs/Engineering Guide/04 AI Coding Tool Compatibility.mdmarkdown3.6 KB
canonical/docs/Engineering Guide/05 MCP Setup.mdmarkdown6.6 KB
canonical/docs/Engineering Guide/Documentation/01 Principles.mdmarkdown15.3 KB
canonical/docs/Engineering Guide/Documentation/02 Structure.mdmarkdown3.8 KB
canonical/docs/Engineering Guide/Documentation/03 HLD Intent.mdmarkdown8.1 KB
canonical/docs/Engineering Guide/Documentation/04 LLD Intent.mdmarkdown9.7 KB
canonical/docs/Engineering Guide/Documentation/05 NFR Intent.mdmarkdown6.0 KB
canonical/docs/Engineering Guide/Documentation/07 Terminology.mdmarkdown1.4 KB
canonical/docs/Engineering Guide/Lifecycles/00 Overview.mdmarkdown5.8 KB
canonical/docs/Engineering Guide/Lifecycles/01 Dev Lifecycle.mdmarkdown19.9 KB
canonical/docs/Engineering Guide/Lifecycles/02 Design Lifecycle.mdmarkdown29.4 KB
canonical/docs/Engineering Guide/Lifecycles/03 Evolve Lifecycle.mdmarkdown7.0 KB
canonical/docs/Engineering Guide/Lifecycles/04 Architecture Lifecycle.mdmarkdown3.6 KB
canonical/docs/Engineering Guide/Lifecycles/05 Documents Flow.mdmarkdown9.5 KB
canonical/manifest.jsonconfig1.6 KB
canonical/mcp.jsonconfig23 B
canonical/root-instructions/INTERFACE.mdmarkdown3.3 KB
canonical/settings.jsonconfig332 B
canonical/skills/design/INTENT.mdmarkdown28.6 KB
canonical/skills/design/TEMPLATE.mdmarkdown19.0 KB
canonical/skills/design/interfaces/architecture.mdmarkdown2.8 KB
canonical/skills/design/interfaces/contract.mdmarkdown7.4 KB
canonical/skills/design/interfaces/design.mdmarkdown2.3 KB
canonical/skills/design/interfaces/efforts.mdmarkdown2.5 KB
canonical/skills/design/interfaces/stack.mdmarkdown1.6 KB
canonical/skills/design/patterns/announce-progress.mdmarkdown997 B
canonical/skills/design/patterns/approval-intelligent.mdmarkdown1.5 KB
canonical/skills/design/patterns/approval-sign-off.mdmarkdown2.4 KB
canonical/skills/design/patterns/context-resolution.mdmarkdown6.6 KB
canonical/skills/design/patterns/design-architect-review.mdmarkdown2.4 KB
canonical/skills/design/patterns/design-oq-triage.mdmarkdown3.8 KB
canonical/skills/design/patterns/design-quality-gate.mdmarkdown1.4 KB
canonical/skills/design/patterns/do-report-commit.mdmarkdown1.2 KB
canonical/skills/design/patterns/docs-writer-authoring.mdmarkdown3.6 KB
canonical/skills/design/patterns/metrics.mdmarkdown11.9 KB
canonical/skills/design/patterns/multi-decision-split.mdmarkdown3.6 KB
canonical/skills/design/patterns/phase-sign-off.mdmarkdown2.4 KB
canonical/skills/design/patterns/satellite-review.mdmarkdown1.5 KB
canonical/skills/design/patterns/session-continuity.mdmarkdown827 B
canonical/skills/design/patterns/task-tracking.mdmarkdown779 B
canonical/skills/design/patterns/waf-review.mdmarkdown4.2 KB
canonical/skills/design/patterns/write-confirm.mdmarkdown963 B
canonical/skills/design/references/00-decision-router.mdmarkdown3.8 KB
canonical/skills/design/references/00-router.mdmarkdown4.6 KB
canonical/skills/design/references/01-context-gathering.mdmarkdown8.1 KB
canonical/skills/design/references/02-ceremony.mdmarkdown4.6 KB
canonical/skills/design/references/03-frame.mdmarkdown6.7 KB
canonical/skills/design/references/04-research.mdmarkdown8.1 KB
canonical/skills/design/references/05-evaluate.mdmarkdown9.0 KB
canonical/skills/design/references/06-solution.mdmarkdown13.7 KB
canonical/skills/design/references/07-review-gate.mdmarkdown4.1 KB
canonical/skills/design/references/08-address-feedback.mdmarkdown8.0 KB
canonical/skills/design/references/09-publish.mdmarkdown8.9 KB
canonical/skills/design/references/10-completion.mdmarkdown7.4 KB
canonical/skills/design/scripts/constants.mjsscript2.0 KB
canonical/skills/design/scripts/design-metrics.mjsscript35.4 KB
canonical/skills/design/scripts/design-metrics.test.mjsscript12.0 KB
canonical/skills/design/scripts/utils.mjsscript3.1 KB
canonical/skills/design/templates/evaluate.mdmarkdown1.4 KB
canonical/skills/design/templates/frame.mdmarkdown2.6 KB
canonical/skills/design/templates/report.mdmarkdown4.8 KB
canonical/skills/design/templates/research.mdmarkdown2.4 KB
canonical/skills/design/templates/stat.jsonconfig1.7 KB
canonical/skills/design/templates/stat.schema.jsonconfig17.1 KB
canonical/skills/dev/INTENT.mdmarkdown15.9 KB
canonical/skills/dev/TEMPLATE.mdmarkdown9.0 KB
canonical/skills/dev/interfaces/intent.mdmarkdown2.6 KB
canonical/skills/dev/patterns/announce-progress.mdmarkdown1.1 KB
canonical/skills/dev/patterns/approval-intelligent.mdmarkdown1.5 KB
canonical/skills/dev/patterns/approval-sign-off.mdmarkdown1.9 KB
canonical/skills/dev/patterns/do-report-commit.mdmarkdown888 B
canonical/skills/dev/patterns/metrics.mdmarkdown17.7 KB
canonical/skills/dev/patterns/phase-sign-off.mdmarkdown2.2 KB
canonical/skills/dev/patterns/review-gate.mdmarkdown3.6 KB
canonical/skills/dev/patterns/satellite-review.mdmarkdown3.1 KB
canonical/skills/dev/patterns/session-continuity.mdmarkdown797 B
canonical/skills/dev/patterns/task-tracking.mdmarkdown766 B
canonical/skills/dev/patterns/write-confirm.mdmarkdown615 B
canonical/skills/dev/references/00-router.mdmarkdown2.2 KB
canonical/skills/dev/references/01-context-gathering.mdmarkdown10.3 KB
canonical/skills/dev/references/02-ceremony.mdmarkdown4.5 KB
canonical/skills/dev/references/03-plan.mdmarkdown14.2 KB
canonical/skills/dev/references/04-implement.mdmarkdown11.7 KB
canonical/skills/dev/references/05-test.mdmarkdown7.8 KB
canonical/skills/dev/references/06-document.mdmarkdown11.7 KB
canonical/skills/dev/references/07-completion.mdmarkdown6.3 KB
canonical/skills/dev/scripts/constants.mjsscript3.2 KB
canonical/skills/dev/scripts/quality.mjsscript9.1 KB
canonical/skills/dev/scripts/quality.test.mjsscript6.6 KB
canonical/skills/dev/scripts/timing.mjsscript38.1 KB
canonical/skills/dev/scripts/timing.test.mjsscript21.0 KB
canonical/skills/dev/scripts/utils.mjsscript3.1 KB
canonical/skills/dev/templates/plan.mdmarkdown3.3 KB
canonical/skills/dev/templates/report.mdmarkdown6.4 KB
canonical/skills/dev/templates/stat.jsonconfig1.4 KB
canonical/skills/dev/templates/stat.schema.jsonconfig29.3 KB
canonical/skills/evolve/INTENT.mdmarkdown1.6 KB
canonical/skills/evolve/TEMPLATE.mdmarkdown1.3 KB
canonical/skills/evolve/interfaces/improvement.mdmarkdown1.2 KB
canonical/skills/evolve/interfaces/reflection.mdmarkdown353 B
canonical/skills/evolve/interfaces/validation.mdmarkdown653 B
canonical/skills/evolve/patterns/task-tracking.mdmarkdown787 B
canonical/skills/evolve/references/00-router.mdmarkdown1.1 KB
canonical/skills/evolve/references/01-reflect.mdmarkdown805 B
canonical/skills/evolve/references/02-triage.mdmarkdown2.6 KB
canonical/skills/evolve/references/03-draft.mdmarkdown1.2 KB
canonical/skills/evolve/references/04-validate.mdmarkdown1.5 KB
canonical/skills/evolve/references/05-adopt.mdmarkdown1.5 KB
canonical/skills/war/INTENT.mdmarkdown4.5 KB
canonical/skills/war/TEMPLATE.mdmarkdown1.8 KB
canonical/skills/war/patterns/present-and-save.mdmarkdown1.0 KB
canonical/skills/war/patterns/scope-resolution.mdmarkdown830 B
canonical/skills/war/references/01-design.mdmarkdown3.4 KB
canonical/skills/war/references/02-implementation.mdmarkdown3.8 KB
canonical/skills/war/templates/design-report.mdmarkdown869 B
canonical/skills/war/templates/implementation-report.mdmarkdown1.8 KB
canonical/skills/where-we-are/INTENT.mdmarkdown3.3 KB
canonical/skills/where-we-are/TEMPLATE.mdmarkdown6.3 KB
canonical/skills/where-we-are/report-template.mdmarkdown2.0 KB
canonical/skills/x-ray/INTENT.mdmarkdown2.4 KB
canonical/skills/x-ray/TEMPLATE.mdmarkdown10.5 KB
implementation-harness.mdmarkdown1.4 KB
package.jsonmanifest696 B
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/CHANGELOG.md.patchother1.2 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/README.md.patchother1.9 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/SKILL.md.patchother2.5 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/canonical/agents/docs-writer/INTENT.md.patchother1.4 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/canonical/docs/.og/cli.test.mjs.patchother846 B
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/canonical/docs/Engineering Guide/04 AI Coding Tool Compatibility.md.patchother3.4 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/canonical/docs/Engineering Guide/05 MCP Setup.md.patchother5.2 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/canonical/manifest.json.patchother662 B
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/canonical/settings.json.patchother150 B
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/canonical/skills/design/INTENT.md.patchother19.4 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/patterns/skill-authoring.md.patchother692 B
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/references/host-capabilities.md.patchother1.2 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/references/launch/01-launch.md.patchother1.1 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/references/launch/01a-launch-plan.md.patchother1.6 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/references/launch/01b-launch-install.md.patchother7.3 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/references/launch/02-launch-questions.md.patchother5.2 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/references/launch/08-provider-launch.md.patchother3.2 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/references/patch/04-patch-harvest.md.patchother2.6 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/references/patch/05-patch-apply.md.patchother3.0 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/references/patch/06-patch-state.md.patchother7.1 KB
patch/20261105T160045Z-3.3.0-copilot-provider-6630df3d-f6bf-4651-8732-8048942ad1fe/diffs/scripts/lib/agent-format.mjs.patchother2.2 KB
# Implementation Harness

The harness is a portable set of canonical skills, agents, and documentation templates. Launch installs a project-specific version of that set. Harvest and
apply keep the canonical sources and permanent project files connected afterward.

**Launch should be as deterministic as possible:** copy the canonical material,
apply known project facts, and derive only the content that needs project-specific
reasoning. The same source version, project evidence, and recorded decisions should
produce the same file selection, paths, substitutions, and required structure.
Derived prose still requires evidence checks; identical wording is not guaranteed.

This document explains the launch contract. JSON materialization checks are executable; semantic evidence and final-file verification remain the orchestrator’s responsibility.

## The canonical/ root

Everything the harness deploys or patches lives under one folder, `canonical/`,
holding `agents/`, `docs/`, `skills/`, provider runtime procedures, and the
`manifest.json`, `settings.json`, and `mcp.json` adapter inputs. It's the single tree `patch
harvest`, `patch prepare`, and `patch apply` read from and write into for portable
content (a narrow, documented set of package-internal exceptions — the package's
own entrypoints, references, changelog, and report template — travel as
`canonical-only` patch entries; see `references/patch/06-patch-state.md`). Every other
top-level item — `references/` (this process's own how-to), `scripts/`
(materializer tooling), `patch/` (immutable bundle history), `patterns/`
(bootstrap-authoring guidance), `reports/`, and the
root docs/entrypoints — is the bootstrap's own scaffolding, not deployable payload.

## Local toolkit and durable project state

Launch first selects one or more of Claude, Codex, and Copilot. The canonical manifest maps sources
to provider destinations; native settings and MCP declarations live beside it.
See [Provider launch](references/launch/08-provider-launch.md) for executable plan generation,
adaptation and verification. Permanent provider files remain editable project artifacts.

Launch answers, plans, receipts and reports live in `.harness/state/launch/` in the project.
The package is read-only and can be replaced at any time; it never holds state. Patch
inventory, baselines and operation records live in `.harness/state/` too. Launch adds
ignore entries for the generated files and never untracks anything. See
`references/patch/06-patch-state.md` for the layout and the commit rule.
Never record secret values.

## Four operations

| Command         | Purpose                                                         | Direction                   |
| --------------- | --------------------------------------------------------------- | --------------------------- |
| `launch`        | Install and adapt the harness for the first time                | Canonical → permanent       |
| `patch harvest` | Promote selected project improvements into reusable sources     | Permanent → canonical       |
| `patch prepare` | Package canonical sources already edited directly into a bundle | Canonical → portable bundle |
| `patch apply`   | Update a project from canonical sources or a supplied patch     | Canonical → permanent       |

`patch` alone asks for the direction. Packaging and recording changes are part of
harvest/prepare/apply, rather than separate modes. `prepare` is optional: `apply`
packages direct canonical edits itself when no `prepare` bundle exists for the
requested scope. Run `prepare` when the bundle needs to exist — for review, or to
carry to another project — before any permanent file is touched.

```mermaid
flowchart LR
    C["Canonical harness<br/>skills · agents · docs"]
    K["Project knowns<br/>evidence · configuration · decisions"]
    P["Permanent project files<br/>installed skills · agents · docs"]
    H["Harvest<br/>select improvement · generalize"]
    PR["Prepare<br/>package direct canonical edits"]
    M["Patch record<br/>manifest · update diffs · additions"]

    K -->|"personalization"| P
    C -->|"launch / apply"| P
    P --> H --> C
    H --> M
    C --> PR --> M
    C -->|"apply records canonical changes"| M

    classDef canonical fill:#15803d,stroke:#166534,color:#fff
    classDef project fill:#2563eb,stroke:#1e40af,color:#fff
    classDef process fill:#7c3aed,stroke:#5b21b6,color:#fff
    classDef record fill:#0f766e,stroke:#115e59,color:#fff
    class C canonical
    class K,P project
    class H,PR process
    class M record
```

Canonical sources own shared behavior. Permanent files contain that behavior adapted
to a project, plus any explicitly retained local customizations. Harvest is the
deliberate promotion of an improvement into the canonical harness.

## Launch: Q&A → JSON → script → INTENT → verify

Launch starts with the material in this package. It does not redesign each skill,
agent, or documentation convention from scratch.

| Treatment  | When to use it                                            | Example                                                                     |
| ---------- | --------------------------------------------------------- | --------------------------------------------------------------------------- |
| Copy       | Content is independent of the project                     | A shared process pattern or documentation companion                         |
| Specialize | A canonical source declares substitutions or conditions   | Ticket prefix, build command, agent scope, enabled lifecycle                |
| Derive     | Actual project evidence is needed to populate the content | A component's responsibilities, integrations, or architecture documentation |

A single file can combine these treatments: preserve its canonical structure,
substitute the known identifiers, and derive its project-specific sections.
Companions and references belong to the same installation unit as their skill.

```mermaid
flowchart TD
    I["Inspect project<br/>code · config · existing docs · existing harness"]
    K["Record knowns<br/>value · evidence · decision status"]
    Q{"Required value<br/>missing or conflicting?"}
    A["Resolve the specific gap<br/>with the developer"]
    P["JSON plan<br/>sources · targets · hashes · knowns · INTENT actions"]
    C["Materialization script<br/>copy canonical files"]
    S["Specialize templates<br/>apply knowns and conditions"]
    D["Apply recorded INTENT actions<br/>derive from project evidence"]
    V["Verify actual files<br/>coverage · evidence · links · local changes"]
    R["Record final state<br/>manifest · baselines · report"]

    I --> K --> Q
    Q -->|"yes"| A --> K
    Q -->|"no"| P
    P --> C --> S --> D --> V --> R

    classDef project fill:#2563eb,stroke:#1e40af,color:#fff
    classDef decision fill:#d97706,stroke:#92400e,color:#fff
    classDef process fill:#7c3aed,stroke:#5b21b6,color:#fff
    classDef verified fill:#15803d,stroke:#166534,color:#fff
    class I,K project
    class Q,A decision
    class P,C,S,D process
    class V,R verified
```

### Applying knowns

Read the repository before asking questions. Record each relevant value with its
source: configuration, code, an existing project decision, or a developer answer.
Distinguish verified facts from proposed interpretations. Resolve conflicting
evidence explicitly; do not silently choose whichever value was read last.

Use established values directly. Ask about missing requirements, ambiguous choices,
or conflicting evidence. Once resolved, reuse the decision throughout the run and
on resume. Unknown values remain in the run/design records; they must not become invented
facts or placeholders in published architecture. A value required for an executable instruction must be resolved before that
instruction is considered ready to use.

### Deriving project documentation

For an existing implementation, derive LLD from code and HLD from verified LLD,
using confirmed project intent where needed. Preserve the distinction between the
implemented state and intended future architecture, particularly during adoption.
For a greenfield project, publish confirmed design statements and keep unresolved
implementation details in the run/design records. An empty LLD directory is not evidence for a populated HLD.

```mermaid
flowchart LR
    C["Existing code and configuration"] --> L["LLD component documents"]
    L --> V["Verify coverage and evidence"] --> H["HLD container documents"]
    Q["Confirmed project intent"] --> H
    G["Greenfield: no implementation"] --> Q
    H --> S["System overview and sizing"]
    C --> N["Testable NFR statements"]
    Q --> N
    S --> R["Review final documentation"]
    N --> R

    classDef evidence fill:#2563eb,stroke:#1e40af,color:#fff
    classDef document fill:#7c3aed,stroke:#5b21b6,color:#fff
    classDef review fill:#15803d,stroke:#166534,color:#fff
    class C,Q,G evidence
    class L,H,S,N document
    class V,R review
```

Architecture uses HLD, LLD, and NFR Intent contracts. Published docs contain supported
statements, meaningful business/data flows and interaction sequences. UI mockups are
included when they clarify user-facing behavior. Questions and missing decisions stay
in working records. NFR statements identify scope, approved targets/controls, operating
conditions, and verification evidence; no arbitrary targets are invented.

## Native tracking and recovery

Available native trackers expose progress and, where supported, dependencies. The dated report and local operation
records preserve progress across sessions. A task is complete when its required
outputs have been verified, not merely when an agent returned a message.

The active provider assigns one agent to each confirmed layer or component. The orchestrator owns shared indexes, cross-document consistency, and final review. Do not model documentation as a separate workflow runtime.

Record the selected inputs, output paths, and outcome of each agent. On interruption,
verify completed outputs and resume unfinished work. A missing agent result is an
incomplete operation. Layer/component agents require implementation evidence; greenfield
work requires confirmed intent. If an agent is unavailable, report the limitation and
choose an explicit fallback before marking its work complete.

## Harvest, apply, and traceability

**Harvest:** compare permanent edits with their recorded baseline, select the reusable
improvements, generalize project values, reconcile against current canonical sources,
then verify and record the change. It does not automatically rewrite permanent files.

**Apply:** select current canonical sources or a supplied patch, reconcile canonical
changes, personalize with this project's knowns, preserve or resolve local edits,
then verify permanent outputs and update their deployment baselines.

- Manifests record operations, paths, rationale, dependencies, and hashes.
- Updated files travel as unified diffs; added files travel in full.
- Renames identify both paths; deletes are explicit operations.
- Update diffs require the expected canonical base or an explicit reconciliation.
- Baseline file contents and project-specific answers stay local to the project.
- Partial work stays partial; successful writes do not conceal failed outputs.

Launch establishes the initial baseline. Later harvest/apply operations use it to
distinguish canonical updates from permanent edits. Record the planned operation
before writes, and finalize baselines only after all intended changes—including
cleanup—have been verified.

## Implementation and verification boundary

`scripts/materialize-launch.mjs` validates and materializes the JSON plan. INTENT-guided
adaptation follows script completion; a materialized receipt does not mean launch is done.
Layer/component agents receive explicit documentation intent and evidence boundaries, return
coverage/evidence, and expose partial failures. The orchestrator verifies actual files.
Canonical INTENT defects are recorded as local plan exceptions, not silently rewritten.

The local materializer checks use an isolated project. Semantic agent behavior still needs a real launch.

## Process references

- [Skill entry point](SKILL.md)
- [Launch recipe](references/launch/01-launch.md)
- [Patch router](references/patch/03-patch.md)
- [Harvest recipe](references/patch/04-patch-harvest.md)
- [Prepare recipe](references/patch/07-patch-prepare.md)
- [Apply recipe](references/patch/05-patch-apply.md)
- [Manifest and state contract](references/patch/06-patch-state.md)

Installed runtime procedures are permanent copies. The bootstrap package may be
removed after verified launch without affecting skill/agent execution.

## Shared package discovery

This repository keeps the authoring source in `skills/implementation-harness/`.
The `.claude/skills/implementation-harness`, `.agents/skills/implementation-harness`, and
`.github/skills/implementation-harness`
symlinks are local discovery links for working on this source repository.

Consumer projects receive a release artifact, not a copy of this authoring directory. A direct
install puts the package in the project's selected host skill directory, such as
`.agents/skills/` for Codex, `.claude/skills/` for Claude, or `.github/skills/` for Copilot; a
marketplace install keeps it in the host plugin cache. The package folder is named
`implementation-harness/`. Record its
physical location as `package:`; all selected hosts then share the project's `.harness/state/`. See
[host capabilities](references/host-capabilities.md) for execution procedures.

## Copilot

Install the package under `.github/skills/` (or through the Copilot plugin marketplace), invoke `/implementation-harness launch`
and select `copilot` in Q6. Claude and Codex may remain selected alongside it.
See [provider destinations](references/launch/08-provider-launch.md#copilot).
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 470kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall