Human-approved coordination

HACP records the authority trail around AI and tool work.

HACP helps humans approve, route, and audit agent work without handing away the decision. It records authority, evidence, custody, and human decisions around AI and tool work.

HACP does not execute work. HACP does not approve outcomes. Human approval remains explicit.

HACP v0.3-candidate

Start with the problem you need to solve.

Eleven evidence-first articles explain the candidate contract for engineering leaders and implementers. Every technical claim links to public source evidence.

Start here What HACP v0.3-candidate Proves and Does Not Prove

The canonical evidence, maturity, and non-claims boundary.

Authority boundary A Receipt Is Evidence, Not Approval

Keep human approval, execution, reports, and receipts distinct.

For agent teams Your Agent Said Done. Who Authorized the Next Step?

A report can inform the next decision. It cannot silently become permission.

All eleven articles Browse the complete HACP v0.3-candidate collection

Choose by evidence, authority, continuation, transport, testing, or stops.

Public proof

Proof is evidence, not approval.

The current public proof is based on HACP repo PR joefeser/hacp#35 and commit 2410d225854fa40354a17655b23df6237cc41616.

The proof shows what happened under a bounded human-approved packet. It does not authorize publication, runtime execution, autonomous continuation, or approval of outcomes.

  • Proof records evidence for a human decision.
  • Proof does not approve the outcome by itself.
  • Proof does not grant the next action.

Operating principle

Nothing is real until the chain is clear.

Nothing is real until the chain of custody, authority, evidence, and approval is clear.

HACP exists to make that chain reviewable for AI and tool work. It keeps evidence separate from approval, records what authority was granted, and makes the next human decision explicit.

Manual approved-loop evidence

The current proof is manual, bounded, and stopped.

Manual approved-loop evidence is not runtime, transport, or self-running evidence.

The public evidence records a manual chain: approved scope, bounded action evidence, returned findings, and an explicit stop before any next site or repo change. That is useful evidence for review. It is not evidence that the loop runs itself.

  • No runtime readiness is claimed.
  • No transport readiness is claimed.
  • No self-running readiness is claimed.

Non-claims boundary

What this proof does not claim.

  • No production readiness claim.
  • No compliance readiness claim.
  • No certification claim.
  • No formal standardization claim.
  • No product completion claim.
  • No loop completion claim.
  • No claim that HACP executes work.
  • No claim that HACP approves outcomes.

Read the bounded pages

Each page keeps evidence separate from authority.

These pages publish the approved site copy for the current public proof and keep the same boundary language across the site.