OpenClaw / Longju Adapter Note

This is a docs-only adapter note.

It does not change runtime behavior, add SACP as a dependency, or claim that OpenClaw or Longju have official SACP support.

Adapter Summary

Project: OpenClaw-style local workspace / Longju local operator
Native concept: local task dispatch, worklog, evidence brief, and `.sacp/` shadow ledger
SACP target: SACP/0.1 receipt plus optional Receipt Chain extensions
Adapter status: docs-only example

This note maps a local agent task or worklog into a SACP receipt so completion claims can be audited for claims, evidence, next owner, and human approval boundary.

Native Flow

The public-safe Longju adoption case used this local flow:

human dispatch -> handoff -> attempt -> evidence -> receipt -> next owner

In an OpenClaw-style workspace, this can remain file-based:

.sacp/
  handoffs/
  attempts/
  evidence/
  receipts/
  snapshots/
  memory-candidates/
  skill-candidates/

SACP does not require this directory layout. It only needs the final receipt to preserve enough state for audit and handoff.

Field Mapping

Native field or concept	SACP field	Notes
human dispatch id / task file path	handoff_id	Stable idempotency key for the work request.
dispatch body hash / source snapshot	source_fingerprint	Used to distinguish duplicate handoff from changed-source rework.
attempt packet	attempt_id	New attempt under the same handoff for retry or reassignment.
local operator name	agent_id	Example: Longju, OpenClawAgent, or a specific worker name.
worklog completion claim	claims[]	Split summary into auditable claims. The worklog is not evidence by itself.
command output / diff / evidence brief	claims[].source_id + verification.evidence_id	Tool and command outputs can support tool_result claims.
runtime guard gate	verification.method	Example: PostTask receipt gate or public-safe local trial review.
.sacp/receipts/<id>.yaml	receipt file path	Storage location is local convention, not core protocol.
next assigned reviewer / human	next_owner	Must be concrete.
release, memory, or skill approval gate	human_decision_required + extensions.sacp.chain.decisions	Durable promotion or risky side effects should require approval evidence.
local checkpoint / context snapshot	extensions.sacp.chain.checkpoint	Pointer to local snapshot or checkpoint; does not change core fields.

Minimal Receipt

protocol: SACP/0.1
type: receipt
method: COMPLETE
status_code: 200
handoff_id: hf_longju_public_safe_trial
attempt_id: attempt_001
agent_id: Longju
claims:
  - text: "The false-completion trial returned 412 missing_evidence."
    claim_type: tool_result
    source_id: ev_public_safe_core_trials
    support_status: supported
  - text: "The duplicate-handoff trial returned 204 no_action_needed."
    claim_type: tool_result
    source_id: ev_public_safe_core_trials
    support_status: supported
verification:
  status: passed
  method: "public-safe local trial review"
  evidence_id: ev_public_safe_core_trials
residual_risk: "This receipt describes one local setup; it does not prove universal agent safety."
next_owner: Human
human_decision_required: false
extensions:
  vendor.openclaw.workspace: local_workspace_path
  vendor.longju.ledger: .sacp
  sacp.chain.profile: sacp-chain
  sacp.chain.project: longju_public_safe_trials
  sacp.chain.module: runtime_guard
  sacp.chain.checkpoint: snapshot_public_safe_core_trials
  sacp.chain.evidence:
    - ev_public_safe_core_trials
  sacp.chain.stop_rule: "Stop before external action, memory promotion, or skill promotion unless approval evidence exists."

Dirty Cases To Test Before Claiming Support

Dirty case	Expected SACP result
Worklog says "done" but no receipt exists	400 invalid_packet
Worklog says "tests passed" with no command output	412 missing_evidence
Same handoff and same source are processed twice	409 duplicate_handoff or link existing receipt
Same handoff arrives with changed source_fingerprint	202 accepted_processing and rework
Attempt expires without a receipt	504 lease_expired and new attempt under same handoff
Memory or skill promotion lacks approval	human_decision_required: true

Claim Boundary Rules

• A worklog summary is not evidence by itself.
• A user instruction remains a user_statement unless external evidence exists.
• Model reasoning remains inference unless retrieved or tool evidence exists.
• Command output, diffs, test output, and structured evidence briefs can support tool_result claims.
• SACP may make a local workflow easier to resume, audit, block, and review; it does not prove the underlying work is correct.

Relationship To The Adoption Case

See ADOPTION_CASE_LONGJU.md for the public-safe adoption case.

This note narrows that case into a reusable adapter mapping. The adoption case reports what one local setup tried; this adapter note shows how similar local records can be represented as SACP receipts.