SACP AgentOps Doctor skill
A skill that audits an agent's done report and shows what is missing: a receipt, test evidence or a human approval.
- 106 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skill reads agent reports with local heuristics and runs or changes nothing.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/sacp-agentops-doctorDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill reads agent reports with local heuristics and runs or changes nothing.
This is third-party code. Review the repository files before installing.
What it does
SACP is an experimental protocol for verifying agents' completion claims, and AgentOps Doctor is a skill built on it. The skill takes raw agent output, a worklog or a handoff and returns a status code, receipt completeness, claim findings, a memory warning, the next owner and the exact fix. Statuses are: 400 for done without a receipt, 412 for a claim with no evidence or approval, 500 for a model error or empty output, and 200 only when the audit finished with no serious findings. The skill does not execute the audited task and does not prove it correct; it uses local heuristics. Next to it sits a Python reference engine that issues a verdict for a gate before a release.
Who it is for. Teams that hand work between agents and want to check phrases like all tests passed before the next step.
Good fit when
- An agent wrote all tests passed and the report has no command output.
- Before publishing a release you need to match test evidence, the artifact and a human approval.
- An agent wants to write output into long-term memory and you want to check that a human confirmed it.
- A handoff between agents has no receipt and you need to know who owns the retry.
Not a fit when
- You need the task itself re-run: the skill does not execute it.
- You need evidence collection: trustworthy collection stays an integration task.
- You need a mature product: the project is labelled an experimental alpha.
Example request
Audit this agent report with AgentOps Doctor and name the exact fix.Limitations
The project is labelled an experimental alpha: a protocol, an engine and examples, while trustworthy evidence collection stays with the integration. The skill uses local heuristics, does not run the original task and does not prove it correct. The reference engine needs Python 3.11 or 3.12. The README has no skill install command through a package manager, so the skill is placed into the agent skills directory by hand. The last commit in the repository was made in October 2026.
How to disable. Delete the `agentops-doctor-skill` folder from the agent skills directory and the SACP clone if it was needed only for the skill.
FAQ
What does AgentOps Doctor return?
A status code, receipt completeness, claim findings, a memory warning, the next owner, the needed human decision, the fix and a translated SACP receipt.
How do I run the check without an agent?
Run `python agentops_doctor.py examples/done_but_no_receipt.md` in the skill folder. Add the `--json` flag for JSON output.
Does the skill accept all tests passed?
Without command output the claim gets status 412: by the skill rule, trust requires a receipt.
What is SACP?
State-Aware Collaboration Protocol: a protocol that records the agent claim, evidence, an external system report and authorization for the next step. A Python engine issues a verdict for a gate.
Does the skill run the audited task?
No. It reads the report and shows what is missing, but does not execute the task or prove it correct.
Discussion
No questions or reviews yet.
Related
SkillsCaliper runs agent tasks with and without your skill, then shows whether the skill helps and how many tokens it costs
A skill and tester agent for Claude Code that test the 1C interface through the 1c-testpilot MCP server with evidence
A CLI and MCP for testing agent skills: repeatable YAML evals, report comparison, regression detection and skilled-improve
A skill and MCP server for the Rhesis platform: the agent designs, runs and analyzes test suites for your own AI agent
A skill that reads a repository once and installs checks: code rules become gates that fail a commit or a CI run.
A skill that helps an agent inspect Scrawl screen recordings using keyframes, spoken explanations and the user’s drawn marks.
README in short
The repository holds the protocol with lifecycle and envelope documents, the `sacp_verify` reference engine with a CLI, reproducible release-check examples and two skills: `agentops-doctor-skill` and a prompt version in `agentops-doctor`. A Chinese README is included.
README badge
Original SKILL.md
# AgentOps Doctor Skill Use AgentOps Doctor when a user gives you messy agent output, worklogs, handoffs, completion reports, or memory claims and asks whether the work is trustworthy. ## Core Behavior Return: - `status_code` - `status_text` - `receipt_completeness` - `claim_findings` - `memory_warning` - `next_owner` - `human_decision_required` - `required_fix` - `translated_receipt` ## Rules - No receipt, no trust. - Do not claim the underlying task was executed. - Do not treat model inference as retrieved fact. - Do not treat user statements as verified facts. - Do not accept "all tests passed" without command output. - Do not promote memory without human approval evidence. - Do not approve publishing without explicit human decision. ## Status Guidance - `400 invalid_packet`: agent says done but no receipt or required fields exist. - `412 missing_evidence`: claim lacks evidence, verification, approval, or tool output. - `500 agent_error`: model/tool failed or returned empty output. - `200 completed`: only when the audit itself completed and no high-risk local finding is detected. ## Output Tone Be concise, operational, and repair-oriented. Always give the exact required fix.