← Back to Skills Marketplace
romainsantoli-web

Firm A2a Bridge

by romainsantoli-web · GitHub ↗ · v2.0.0
cross-platform ⚠ suspicious
439
Downloads
0
Stars
1
Active Installs
2
Versions
Install in OpenClaw
/install firm-a2a-bridge
Description
Agent-to-Agent (A2A) Protocol RC v1.0 bridge for OpenClaw agents. Enables agent discovery, identity cards, task lifecycle management, push notification confi...
README (SKILL.md)

firm-a2a-bridge

⚠️ Contenu généré par IA — validation humaine requise avant déploiement en production.

Purpose

Ce skill implémente le A2A Protocol RC v1.0 (Agent-to-Agent) pour OpenClaw, permettant aux agents de se découvrir mutuellement, d'échanger des tâches, et de recevoir des notifications push — tout cela via une interface MCP standardisée.

Gaps couverts :

Gap Sévérité Outil
G1 — Pas de génération Agent Card CRITICAL openclaw_a2a_card_generate
G2 — Pas de validation Agent Card CRITICAL openclaw_a2a_card_validate
G3 — Pas d'envoi de tâches A2A HIGH openclaw_a2a_task_send
G4 — Pas de suivi de tâches A2A HIGH openclaw_a2a_task_status
G5 — Pas de push notifications A2A MEDIUM openclaw_a2a_push_config
G6 — Pas de découverte d'agents HIGH openclaw_a2a_discovery
G7 — Pas d'annulation de tâches HIGH openclaw_a2a_cancel_task
G8 — Pas de souscription SSE HIGH openclaw_a2a_subscribe_task

Tools

openclaw_a2a_card_generate

Génère un fichier .well-known/agent-card.json à partir d'un fichier SOUL.md. Extrait automatiquement l'identité, les compétences, les capacités et les schémas de sécurité depuis le frontmatter et le corps du SOUL.md.

Paramètres :

  • soul_path (str, required) — Chemin vers le fichier SOUL.md
  • base_url (str, required) — URL de base où l'agent sera joignable
  • output_path (str, optional) — Chemin de sortie pour le JSON
  • capabilities (dict, optional) — Capacités A2A (streaming, pushNotifications)
  • security_schemes (dict, optional) — Schémas OAuth2/apiKey/http
  • extensions (list, optional) — Déclarations d'extensions A2A
  • sign (bool, optional) — Signer la carte avec JCS+JWS
  • signing_key (str, optional) — Clé de signature (masquée dans l'output)
  • default_input_modes (list, optional) — Types MIME d'entrée par défaut
  • default_output_modes (list, optional) — Types MIME de sortie par défaut

Exemple :

{
  "name": "openclaw_a2a_card_generate",
  "arguments": {
    "soul_path": "./souls/ceo/SOUL.md",
    "base_url": "https://agents.example.com/ceo"
  }
}

openclaw_a2a_card_validate

Valide un Agent Card contre la spécification A2A v1.0 RC. Vérifie les champs requis, le format des URLs, la structure des skills, les capabilities, et les schémas de sécurité.

Paramètres :

  • card_path (str, optional) — Chemin vers un fichier agent-card.json
  • card_json (dict, optional) — Agent Card inline (alternative à card_path)

openclaw_a2a_task_send

Envoie un message/tâche à un agent A2A distant. Crée un Task dans le lifecycle A2A (submitted → working → completed/failed). Inclut une protection SSRF contre les URLs localhost.

Paramètres :

  • agent_url (str, required) — URL de l'agent cible
  • message (str, required) — Message texte à envoyer
  • context_id (str, optional) — ID de contexte pour grouper les tâches
  • blocking (bool, optional) — Attendre la complétion
  • metadata (dict, optional) — Métadonnées additionnelles

openclaw_a2a_task_status

Récupère le statut d'une tâche ou liste les tâches. Mappe les opérations GetTask / ListTasks du protocole A2A.

Paramètres :

  • task_id (str, optional) — ID de tâche spécifique (GetTask)
  • context_id (str, optional) — Filtre par contexte (ListTasks)
  • include_history (bool, optional) — Inclure l'historique des messages

openclaw_a2a_push_config

CRUD pour les configurations de webhook push notification A2A. Supporte create, get, list, delete avec protection SSRF.

Paramètres :

  • task_id (str, required) — ID de la tâche concernée
  • action (str) — create, get, list, delete
  • webhook_url (str, optional) — URL du webhook (requis pour create)
  • auth_token (str, optional) — Token Bearer pour la livraison
  • config_id (str, optional) — ID de config (requis pour get/delete)

openclaw_a2a_cancel_task

Annule une tâche A2A en cours d'exécution (CancelTask RC v1.0). Erreur si la tâche est déjà en état terminal.

Paramètres :

  • task_id (str, required) — ID de la tâche à annuler

openclaw_a2a_subscribe_task

Souscrit aux mises à jour d'une tâche via SSE (SubscribeToTask RC v1.0). Diffuse TaskStatusUpdateEvent et TaskArtifactUpdateEvent.

Paramètres :

  • task_id (str, required) — ID de la tâche à surveiller
  • callback_url (str, optional) — URL de callback pour les events

openclaw_a2a_discovery

Découvre les agents A2A via leurs endpoints Agent Card ou scan local de SOUL.md.

Paramètres :

  • urls (list[str], optional) — URLs d'agents à scanner
  • souls_dir (str, optional) — Répertoire local avec des fichiers SOUL.md
  • check_reachability (bool, optional) — Vérifier l'accessibilité des URLs

Architecture

SOUL.md files                    A2A Protocol v1.0 RC
     │                                  │
     ├── card_generate ──► .well-known/agent-card.json
     │                                  │
     ├── card_validate ◄── Spec validation (8 checks)
     │                                  │
     ├── task_send ────────► SendMessage (submitted → working → completed)
     │                                  │
     ├── task_status ──────► GetTask / ListTasks
     │                                  │
     ├── cancel_task ──────► CancelTask (→ canceled)
     │                                  │
     ├── subscribe_task ───► SubscribeToTask (SSE events)
     │                                  │
     ├── push_config ──────► CRUD webhook configs (SSRF-safe)
     │                                  │
     └── discovery ────────► Agent Card endpoint probe + local SOUL.md scan

Security

  • Path traversal protection : tous les chemins de fichiers validés par Pydantic
  • SSRF protection : URLs localhost/127.0.0.1/0.0.0.0/::1 bloquées pour task_send et push_config
  • URL scheme validation : seuls http/https acceptés
  • Token masking : les auth_token sont masqués (4 derniers chars visibles)
  • Input validation : tous les paramètres validés par Pydantic v2 avec contraintes strictes

Testing

python -m pytest tests/test_smoke.py -k "TestA2aBridge" -v

17 tests couvrent :

  • Génération de carte depuis un SOUL.md valide (avec signature JCS+JWS)
  • Gestion des fichiers manquants
  • Protection path traversal
  • Validation de cartes valides/invalides (détection patterns v0.4.0 dépréciés)
  • Envoi de tâches avec SSRF protection
  • Annulation de tâches (CancelTask)
  • Souscription SSE (SubscribeToTask)
  • Statut de tâches (GetTask/ListTasks)
  • Configuration push notifications (CRUD + SSRF)
  • Découverte locale d'agents
Usage Guidance
This skill appears to do what it says (A2A bridge) but lacks provenance (no homepage; unknown source) and omits critical security details. Before installing or using it: 1) Ask the author for an implementation reference (repo or docs) and a maintainer identity. 2) Request specifics about the SSRF protections (what URL patterns are blocked, how localhost/internal addresses are handled). 3) Ask how signing keys and webhook auth tokens should be provided and stored (avoid pasting secrets into free-text fields; prefer use of platform-provided secret stores). 4) Clarify safe defaults for discovery (do not allow an unbounded filesystem scan; require an explicit souls_dir). 5) Test in an isolated non-production environment first, and monitor outgoing network calls to ensure callbacks do not exfiltrate sensitive data. If these clarifications are not available, treat the skill as higher-risk and avoid granting it network access or secrets.
Capability Analysis
Type: OpenClaw Skill Name: firm-a2a-bridge Version: 2.0.0 The skill is classified as suspicious due to its extensive capabilities involving network communication and local file system access, which inherently carry high risks. Specifically, tools like `openclaw_a2a_task_send`, `openclaw_a2a_push_config`, `openclaw_a2a_subscribe_task`, and `openclaw_a2a_discovery` (all described in SKILL.md) perform network requests to arbitrary URLs, while `openclaw_a2a_card_generate` and `openclaw_a2a_card_validate` access local files. Although the `SKILL.md` documentation claims robust security measures such as SSRF and path traversal protection, these are claims without code verification. A vulnerability in the implementation of these protections could lead to critical issues like data exfiltration or arbitrary file access, classifying it as suspicious rather than benign, but without clear evidence of intentional malicious behavior.
Capability Assessment
Purpose & Capability
The name and description (A2A bridge for agent discovery, task lifecycle, push notifications, SSE) align with the SKILL.md tools and parameters (card generation/validation, send/status/cancel tasks, push config, discovery). The declared scope matches the functionality described.
Instruction Scope
Instructions require reading SOUL.md files and optionally scanning a local directory of SOUL.md files, and they permit sending HTTP requests/webhooks/SSE to arbitrary agent URLs and callback URLs. These behaviors are consistent with the purpose but are security-sensitive. The doc claims SSRF protection without describing how it is enforced, and it does not limit or document safe defaults for local directory scans (souls_dir) or network targets — leaving room for accidental exposure or misuse.
Install Mechanism
This is an instruction-only skill with no install spec and no declared binaries, which is low-risk from an installation perspective (nothing is written to disk by an installer).
Credentials
No environment variables or credentials are required, which is proportional. However the tool exposes parameters for a signing_key and auth_token (for webhook delivery) without guidance about where or how to provide/store them securely. The SKILL.md also promises that signing keys are 'masked in the output' but gives no implementation or storage details.
Persistence & Privilege
The skill does not request always:true and does not declare any persistent config paths or system-wide modifications. Autonomous invocation is allowed (default), which is normal for skills.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install firm-a2a-bridge
  3. After installation, invoke the skill by name or use /firm-a2a-bridge
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v2.0.0
A2A RC v1.0 rewrite: 8 tools, cancel, subscribe SSE, push CRUD, JWS signing
v1.0.0
Initial release of firm-a2a-bridge (v1.0.0): - Implements A2A Protocol v1.0 RC bridge for OpenClaw agents, covering gaps G1-G6 from the Phase 7 roadmap. - Enables agent discovery, agent-card generation/validation, task lifecycle management, and push notification configuration per the A2A specification. - Provides tools: card generation from SOUL.md, card validation, remote task send/status, push notification CRUD, and agent discovery. - Includes robust security features: path traversal and SSRF protection, strict input validation, and token masking. - Comprehensive tests for all major functionalities and security protections.
Metadata
Slug firm-a2a-bridge
Version 2.0.0
License
All-time Installs 1
Active Installs 1
Total Versions 2
Frequently Asked Questions

What is Firm A2a Bridge?

Agent-to-Agent (A2A) Protocol RC v1.0 bridge for OpenClaw agents. Enables agent discovery, identity cards, task lifecycle management, push notification confi... It is an AI Agent Skill for Claude Code / OpenClaw, with 439 downloads so far.

How do I install Firm A2a Bridge?

Run "/install firm-a2a-bridge" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Firm A2a Bridge free?

Yes, Firm A2a Bridge is completely free (open-source). You can download, install and use it at no cost.

Which platforms does Firm A2a Bridge support?

Firm A2a Bridge is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Firm A2a Bridge?

It is built and maintained by romainsantoli-web (@romainsantoli-web); the current version is v2.0.0.

💬 Comments