agentleFS
Sign inStart free
bks-labbks-lab/open-bridge/skills/a2a/SKILL.md

a2a

Skill13 starsChanged 13 days agoMITView on GitHub

Talks to other Bridges and A2A agents: ask a peer, check an agent card, probe an endpoint''s auth and topic boundary, and scaffold an A2A agent (public, or a peer endpoint with trust: peer). Peers live in infra/a2a-peers/. Trigger: "a2a", "ask <name>''s bridge", "agent card", "check the card", "probe the agent", "new a2a agent", "peer endpoint".

6 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the referenced file ONLY when the branch below sends you there.

    line 23

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

a2a

One skill for both directions of A2A: this Bridge calling another agent, and this Bridge offering an agent of its own. Protocol mechanics only. Who your peers are, which topics they may ask and in which language lives in data: infra/a2a-peers/, the agent's own agent.yaml and system-prompt.md, and an org overlay's skill on top (for example a <org>-a2a skill with the org's peer list and probe prompts).

Engine: skills/a2a/scripts/a2a.py behind skills/a2a/a2a.sh. Stdlib plus PyYAML. Read the referenced file ONLY when the branch below sends you there.

Commands

CommandDoes
peerslist infra/a2a-peers/*.yaml
card <peer|url>fetch the card, report endpoint, version, dialect, https, skills, schemes
ask <peer> "<question>" [--context ID] [--wait S] [--json]send one question in the peer's dialect; --wait polls a held task (a peer waiting for its owner's approval)
get <peer> <task_id> [--json]read a held task again (approved meanwhile?) without sending anything new
request <peer> "<what>" [--subject S] [--wait S] [--json]ask the peer's OWNER to do something; the owner (or a rule the owner set) decides, the model never sees it. Needs owner_request on the card
rules <peer> [--json]the owner's rules for you and your request history at that peer (owner_policy)
probe <peer> [--prompts FILE]no token and wrong token must be refused, then each boundary prompt is checked against its expected refusal or answer
new-agent <name> --trust public|peer [--port N] [--peer ID:ENV:LOGIN]scaffold agents/<name>/; peer writes agent.yaml with auth, a prompt skeleton with a topic boundary, share/ and launch.sh

Exit codes: 0 clean, 1 a finding or a task that did not complete, 2 usage or a refused input.

The token never passes through you

A peer file names its token as a secret reference (credential_ref: keychain://...). ask and probe re-run themselves under skills/secrets/secrets.sh run --env A2A_TOKEN=<ref>, so the value goes into the child process and nowhere else. Never ask the user for a token in chat, never put one in a command line.

Decision tree

What does the user want?
├── Ask another Bridge something          → `a2a.sh ask <peer> "..."`; unknown peer →
│                                           `a2a.sh peers`, then references/peers.md
├── Ask another Bridge's owner to DO sth  → `a2a.sh request <peer> "..." --subject a/b/c`,
│                                           then `get`; their rules: `a2a.sh rules <peer>`
├── Let peers request actions from you    → references/peer-agent.md § 4b
├── Per-peer folders, topics on request  → references/peer-agent.md § 4c
├── "Is that agent set up right?"         → `a2a.sh card <peer|url>`, then `probe`
├── Add a peer to call                    → references/peers.md
├── Offer an agent to the public          → `new-agent <name> --trust public`, then
│                                           references/public-agent.md
├── Offer an agent to known Bridges only  → `new-agent <name> --trust peer --peer ...`,
│                                           then references/peer-agent.md
├── Wire details, 1.0 vs 0.3, errors      → references/protocol.md
├── Something behaves oddly               → references/traps.md
└── Identity, delegation, where it goes   → references/security-model.md

Files

FileHolds
scripts/a2a.pythe engine
assets/peer-*.tmpl, assets/share-README.mdwhat new-agent --trust peer writes
assets/probe-prompts.yamlgeneric boundary prompts; an overlay ships its own set
tests/test_a2a.pyhermetic tests, bash skills/a2a/run-tests.sh
../../infra/a2a-peers/_template.yaml, _schema.yamlone peer

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.

Check your setup, free