Authority packets
Bounded, human-approved work records that define scope and limits.
Working draft protocol
Human approval gates for AI agent workflows.
HACP is a draft protocol for bounded agent delegation, reviewable handoffs, decision gates, and audit evidence.
01
Authority packet
02
Agent report
03
Human decision record
04
Audit evidence
Problem
Agents can do real work, but authority often gets blurred.
AI agents can draft, call tools, communicate, and execute tasks. Most workflows do not clearly separate request, authority, approval, execution, and evidence. That makes it harder to know what was approved, what happened, and which human decision should come next.
What HACP provides
Agent reports
Structured return evidence from agents, adapters, tools, or workflows.
Review findings
Reviewable observations that keep risks and boundary questions visible.
Human decision records
Explicit human decisions after review, separate from agent output.
Audit evidence
Records that help teams inspect the handoff and decision trail.
Transport boundaries
Clear separation between moving work records and granting authority.
Example flow
- Human request
- Authority packet
- Agent work
- Agent report
- Human review
- Approve/reject
- Next workflow step
Who it is for
Enterprise AI teams
Define reviewable boundaries for agent-assisted work.
Agent framework builders
Expose handoff and evidence records without owning every workflow.
Compliance / audit reviewers
Inspect what was approved, returned, reviewed, and decided.
Internal tools teams
Route agent outputs through explicit human decision gates.
Read the draft