Context engineering for AI powered
delivery.

Harness Intelligence shows the work around the diff: the context that shaped the agent, the workflow it followed, the checks that matched the change, and the evidence left for the next developer.
entry points
View operating loop
punks
user@harness-intelligence ~/repos/harness-intelligence (main) 

AGENTS.md

Agent definitions, capabilities and tool contracts.
Context

PLAN.md

Execution plan, steps, decisions and assumptions.
Plan

validation

Automated checks, test results and quality gates.
Validation

report

Run summary, artifacts and key outcomes.
Report
Start with the repository

Scaffold the operating surface.

Use punks scaffold setup to detect the repo, resolve the right packs, and write the prompt, skill, hook, and manifest surfaces the next agent should follow.
punks scaffold setup
punks scaffold setup
 
What the repo gets

Shared context. Visible quality bar.

Harness Intelligence connects rules, workflows, and checks to your reality, so every change is measured, reviewed, and improved.
01

Context

rules and memory
/docs/wiki/harness
.agents/notes
Project context
02

Workflows

phase contracts
Plan
Build
Review
Ship
03

Artifacts

inspectable trail
plan.md
review.md
checks.json
trace.log
04

Validation

surface-matched checks
Spec <-> Implemented
Docs <-> Behavior
Tests <-> Requirements
05

Adoption

baseline feedback
Quality bar 82%
Trend 7d
Report shipped
The operating loop

Intent becomes evidence.

Each step has a contract, an artifact, and a closeout signal, so the next developer can inspect how the work moved from request to evidence.
  1. 01
    Intent
    SPEC.md
  2. 02
    Scope
    PLAN.md
  3. 03
    Plan
    PLAN.md
  4. 04
    Implementation
    diff
  5. 05
    Validation
    checks
  6. 06
    Review
    review.md
  7. 07
    Docs / wiki / backlog / report
    docs
How to use it

Choose the workflow that matches the risk.

  1. 01
    Adopt a repository
    Initialize the requirements gate and wiki tree before repo-aware setup.
    $ punks scaffold init
  2. 02
    Run a manual workflow
    Invoke the implementation skill against a reviewed spec and plan.
    $ codex -p '$implement-spec  implement apps/wiki/specs/<domain>/<spec>  in sequential mode; read SPEC.md and PLAN.md first;  update PLAN.md and IMPLEMENTATION-NOTES.md after each task.'
  3. 03
    Use an AFK phase
    Start a bounded goal-mode bug discovery run with explicit scope and proof output.
    /goal Use $bug-discovery-phase for this repo.Scope: apps/wiki. Provider: codex. Model: gpt-5.Run ClawPatch with --limit 25 --jobs 4.Report: .clawpatch/reports/wiki-discovery.md.OK
  4. 04
    Inspect the evidence
    Review findings, trace evidence to sources, and export what you need for your gate.
    risk-report.jsonJSON
    evidence-bundle.zipZIP
    sbom.spdx.jsonJSON

Keep going from docs.

The docs index is the entrypoint for work flows, trust model, validation gates, and developer runbooks.
Entry points
Trust model
Validation gates
Runbooks
$ hi docs index> Opening documentation index...> Ready.