avp
Skill9 starsChanged 33 days agoApache-2.0View on GitHub
Set up and use AVP (Agent Voyager Project) to run agents, configure Commissions, compare agents or prompts with evaluations, and inspect trajectories. Use when onboarding to AVP, debugging an AVP run, or adapting an existing agent runtime to the Commission, Agent Descriptor, and Trajectory specs.
In this repository2 other files it gives its agents.
- AGENTS.mdAGENTS.md
- writing-guidelinesSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 2 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Read the spec overview before constructing wire objects.
- Very specific: line 23
Use avp agent list, ping, and describe to inspect the installed agent.
- Very specific: line 28
Write a Commission file following the Commission spec.
- Very specific: line 31
Run avp run --agent <name> --commission <file>.
- Distinctive: line 39
Read the spec rather than guessing unsupported settings.
- Very specific: line 61
Test native event translation into a Trajectory, install the adapter's image, check ping and describe, and run avp conformance check.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Agent Voyager Project
A supervisor sends one Commission; the agent's existing runtime does the work and streams its Trajectory. There is no mid-run supervisor push channel. Read the spec overview before constructing wire objects.
Run an agent
- Inspect the installed CLI's
--help. The rebuilt v0.1 CLI currently needs a repository checkout and uses OpenSandbox with a Docker-compatible container engine. From the checkout, useuv run --project avp-cli avp. - Install an adapter image with
avp agent install <name> <Dockerfile> --context <repository-root>. SDK dependencies are still local to the repository. Useavp agent list,ping, anddescribeto inspect the installed agent. - Store a model key with
avp vault set <name>, which prompts locally. Never request secret values in chat or write them into a Commission. Reference it asmodel.credential: {"proxy": "<name>"}. For environment references, pass the named variable with--env NAMEor--env-file. - Write a Commission file following the Commission spec.
Required fields are
avp_version,run_id,model, andprompt. The model is an object withproviderandid. Choose a fresh run ID for each dispatch. - Run
avp run --agent <name> --commission <file>. Real provider calls cost money. Report the outcome and the saved Trajectory location. Inspect errors before retrying a failed run.
Agent-specific built-in settings live under agents, keyed by the Descriptor's
agent_name. Omitted allowlists allow all built-ins; [] allows none. The
Commission adds its own MCP servers and skill files alongside built-ins.
system_prompt replaces the runtime's prompt. output_schema describes an object.
Read the spec rather than guessing unsupported settings.
Evaluate and inspect
In the checkout, read avp-cli/docs/eval.md for the current eval format. Evals
specify eval_version, name, commission, dataset, and scorer, and select
installed agents in the file or through --agent. The built-in scorers are
match and command. Use avp eval run, list, and show to run comparisons
and inspect results. Saved Commissions and the TUI are not implemented yet.
Compare correctness, cost, and response counts separately. Preserve the run's Commission and Trajectory as evidence. Reduce per-response usage and cost once; do not add runtime totals or count subagent usage twice. See the Trajectory spec.
Integrate a runtime
In a checkout, read avp-sdk/README.md and a migrated adapter's README. The SDK
provides types, Recorder, run plan, pricing, endpoints, and agent command line.
The runtime still owns its loop. Describe only its actual built-in capabilities,
using the Agent Descriptor spec.
Test native event translation into a Trajectory, install the adapter's image,
check ping and describe, and run avp conformance check. Report passes and
capability skips together. Use the adapter's real-model checks after changes to
runtime integration. Keep unsupported behavior explicit in its Known gaps.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

