synapse-a2a

name	synapse-a2a
license	MIT
description	Synapse A2A agent communication -- sending messages, spawning agents, delegating tasks, sharing memory, managing the LLM wiki, and coordinating file edits. Use this skill when: running synapse send/reply/broadcast/interrupt, spawning agents with synapse spawn or synapse team start, sharing knowledge with synapse memory, managing wiki pages with synapse wiki, locking files with synapse file-safety, checking agent status with synapse list/status, or orchestrating any multi-agent workflow. For AI/programmatic use, prefer synapse list --json, synapse status <target> --json, or the MCP list_agents tool instead of interactive synapse list.

Synapse A2A Communication

Inter-agent communication framework via Google A2A Protocol.

Worktree Discipline (subagents, read this first)

NEVER cd into .synapse/worktrees/<name>/ directories.

Subagents (Claude Code Agent tool, Codex subprocess, and any other sub-process driven by the parent session) inherit a persistent shell from the parent. A stray cd into a worktree leaks out of the subagent turn and silently corrupts the parent's working directory — git status, git diff, and even git commit then land on the wrong worktree, which wastes debugging time and can put commits on the wrong branch.

Rules for working with .synapse/worktrees/:

• Do not cd into a worktree, ever. Stay in the original working directory for the entire session.
• Read and write files inside a worktree using absolute paths only (for example Read /Volumes/.../.synapse/worktrees/foo/src/bar.py, not cd .synapse/worktrees/foo && cat src/bar.py).
• Run git against a worktree with git -C /abs/path/to/worktree ... instead of changing directory.
• Worktrees are managed by Synapse (synapse spawn --worktree, synapse team start --worktree). Treat them as read/write data surfaces, not as places to live.

If you need to operate from inside a worktree (e.g. running pytest there), spawn a dedicated agent for it with synapse spawn --worktree rather than changing the parent shell's directory.

Quick Reference

Task	Command
List agents	synapse list for humans (auto-refresh, interactive: arrows/1-9 select, Enter jump, k kill, / filter). For AI/scripts use synapse list --json, synapse list --plain, or MCP list_agents
Agent detail	synapse status <target> [--json]
Stuck-agent watchdog (Stage 1)	synapse watchdog check [--alarm-only] [--json] (one-shot heuristic scan; #646)
Send message	synapse send <target> "<msg>" (default: --notify; --from auto-detected)
Broadcast	synapse broadcast "<msg>"
Wait for reply	synapse send <target> "<msg>" --wait
Fire-and-forget	synapse send <target> "<msg>" --silent
Reply	synapse reply "<response>"
Reply to specific	synapse reply "<response>" --to <sender_id>
Reply with failure	synapse reply --fail "<reason>"
Interrupt (priority 4)	synapse interrupt <target> "<msg>"
Send keys to PTY (escape hatch for TUI dialogs; #695)	synapse send-keys <target> <keys> (e.g. a for codex "don't ask again", \r for Enter; bypasses A2A — use when an agent is stuck on an interactive dialog without synapse jump)
Spawn agent	synapse spawn <type> --name <n> --role "<r>" -- <tool-specific-automation-args>
Spawn + send first task (preferred for delegation)	synapse spawn <type> --name <n> --role "<r>" --task-file <path> --task-timeout 600 --notify
Spawn with worktree	synapse spawn <type> --worktree --name <n> --role "<r>" -- <tool-specific-automation-args>
Team start	synapse team start <homogeneous-profiles...> [--worktree] -- <tool-specific-automation-args>

Collaboration Decision Framework

Evaluate collaboration opportunities before starting work:

Situation	Action
Small task within your role	Do it yourself
Task outside your role, READY agent exists	Delegate: synapse send --notify or --silent
No suitable agent exists, need to delegate a task	Spawn + task in one command: synapse spawn <type> --name <n> --role "<r>" --task-file <spec.md> --task-timeout 600 --notify. This spawns, waits for READY, and sends the first task — no manual readiness polling needed.
Need a bare agent (no initial task)	synapse spawn <type> --name <n> --role "<r>" (send tasks later via synapse send)
Stuck or need expertise	Ask: synapse send <target> "<question>" --wait
Completed a milestone	Report: synapse send <manager> "<summary>" --silent
Discovered a pattern	Share: synapse memory save <key> "<pattern>" --tags ... --notify

Recommended Collaboration Gate (3+ phases OR 10+ file changes): Consider these steps before diving into large work:

1. synapse list --json or MCP list_agents — check available agents
2. synapse memory search "<topic>" — check if someone already solved this
3. Build Agent Assignment Plan (Phase / Agent / Rationale) when delegation is beneficial
4. Spawn specialists if needed (prefer different model types for diversity)

Skip this gate for small/medium tasks where the overhead exceeds the benefit.

Use Synapse Features Actively

Feature	Why It Matters	Commands
Shared Memory	Collective knowledge survives agent restarts	synapse memory save/search/list
File Safety	Locking prevents data loss when two agents edit the same file -- skip inside worktrees (SYNAPSE_WORKTREE_PATH)	synapse file-safety lock/unlock/locks
Worktree	File isolation eliminates merge conflicts in parallel editing	synapse spawn --worktree
Broadcast	Team-wide announcements reach all agents instantly	synapse broadcast "<msg>"
History	Audit trail tracks what happened and when	synapse history list/show/stats
Probabilistic Recall	Recall relevant past task observations by recency, importance, and keyword overlap without dumping all history	HistoryManager.recall_observations
Plan Approval	Gated execution ensures quality before action	synapse approve/reject
Canvas	Visual dashboard for sharing rich cards and templates (briefing, comparison, dashboard, steps, slides, plan); cards downloadable as Markdown, JSON, CSV, or native format via browser button or GET /api/cards/{card_id}/download	synapse canvas post/link/briefing/plan/open/list/restart
Agent Control	Browser-based agent management via Canvas #/admin view (select agents, send messages, view responses, double-click agent row to jump to terminal)	synapse canvas open → navigate to #/admin
Workflow View	Browser-based workflow management via Canvas #/workflow view (list workflows, inspect steps, create/edit/delete/import/export workflow YAML, trigger runs, monitor progress with live SSE updates; run history persisted to SQLite across restarts)	synapse canvas open → navigate to

When to Use Canvas

Use Canvas when the output benefits from visual structure or will be referenced later:

• Use Canvas for: diagrams, comparison tables, multi-step plans, design docs, results with rich formatting
• Skip Canvas for: simple completion reports, single-file changes, quick status updates (use broadcast or reply instead)

Template selection guide:

• briefing — structured reports, status updates, release summaries
• comparison — before/after, option trade-offs, review diffs
• steps — plans, migration sequences, execution checklists
• slides — walkthroughs, demos, page-by-page narratives
• dashboard — multi-widget operational snapshots, compact status boards
• plan — task DAGs with Mermaid visualization and step tracking

Use raw synapse canvas post <format> for single blocks; templates for multi-section content.

Spawning Decision Table

⚠️ Same-model rule — try subagents first. When a Claude Code agent needs another claude (or a codex agent needs another codex), use the in-process subagent (Agent / Task tool for Claude, subprocess for Codex) before reaching for synapse spawn. Spawning the same model on the same account shares the rate-limit window — it doubles consumption against the same quota instead of distributing it. Reserve same-model synapse spawn for cases where the helper must outlive the parent session, needs file isolation that subagents can't provide, or holds a distinct long-running role.

synapse spawn is the right tool for cross-model delegation (Claude → codex / gemini), agents that lack subagent support (Gemini / OpenCode / Copilot), or persistent multi-task helpers.

Default spawn policy: When using synapse spawn, pass the underlying CLI's tool-specific automation args after -- so spawned agents can run unattended. For most CLIs this is an approval-skip / auto-approve flag; for OpenCode use --agent build to select the build agent profile and rely on OpenCode's permission config for approval behavior.

Apply the same rule to synapse team start: include the appropriate forwarded CLI args by default, and keep teams homogeneous when those args are CLI-specific.

Common defaults (Synapse already injects these automatically — pass --no-auto-approve to opt out):

• Claude Code: synapse spawn claude --name <n> --role "<r>" -- --permission-mode=auto
• Gemini CLI: synapse spawn gemini --name <n> --role "<r>" -- --approval-mode=yolo
• Codex CLI: synapse spawn codex --name <n> --role "<r>" (synapse injects -cdefault_permissions=":workspace"; Codex 0.128+ removed --full-auto)
• OpenCode: synapse spawn opencode --name <n> --role "<r>" -- --agent build (selects the build agent profile; not a skip-approval flag)
• Copilot CLI: synapse spawn copilot --name <n> --role "<r>" -- --allow-all
• Claude team: synapse team start claude claude -- --permission-mode=auto
• Gemini team: synapse team start gemini gemini -- --approval-mode=yolo
• Codex team: synapse team start codex codex (synapse injects -cdefault_permissions=":workspace")
• OpenCode team: synapse team start opencode opencode -- --agent build (selects the build agent profile; permission prompts still depend on OpenCode config)
• Copilot team: synapse team start copilot copilot -- --allow-all

2026-04 migration: Anthropic deprecated --dangerously-skip-permissions in favor of --permission-mode=auto (safety classifier instead of disabling all checks). Gemini similarly recommends --approval-mode=yolo over the legacy --yolo / -y short forms. Synapse now injects the new flags by default; the legacy forms still work and remain in each profile's alternative_flags.

Condition	Action
Existing READY agent can handle it	synapse send — reuse is faster (avoids startup overhead)
Same-model helper needed (Claude → claude, Codex → codex)	Use the in-process subagent first (Agent/Task tool for Claude, subprocess for Codex). synapse spawn same-model shares the rate-limit window.
Need parallel execution	synapse spawn with --worktree -- <tool-specific-automation-args> for file isolation (cross-model preferred)
Task needs a different model's strengths	synapse spawn a different type (Claude spawns Gemini / Codex, etc.)
User specified agent count	Follow exactly
Single focused subtask	Subagent (same model) or synapse spawn (cross model)
N independent subtasks	Subagents for same-model fan-out, synapse spawn for cross-model

Spawn lifecycle (preferred, one-command): synapse spawn --task-file ... --task-timeout 600 --notify → wait for A2A completion notification → evaluate result → synapse kill <name> -f → confirm in synapse list --json

Legacy lifecycle (only when you need control between spawn and first task): spawn → poll synapse list --json or synapse status <target> --json for READY (allow several minutes; default 30s timeout is too short for most profiles) → synapse send --notify → evaluate → synapse kill -f → confirm cleanup.

⚠️ Common pitfall: sending to an agent that is not yet READY either hangs at the HTTP layer or blocks on the internal readiness wait. Either use synapse spawn --task-file (preferred — it handles readiness for you), or explicitly confirm "status": "READY" before calling synapse send. Do not assume 30 seconds is enough — most profiles take 1-5 minutes.

Agent status set (synapse list --json .status):

Status	Meaning	Action
READY	Idle, can accept new work	synapse send
SENDING_REPLY	Temporarily sending an outbound A2A send/reply POST	Wait; previous status is restored after the POST finishes
PROCESSING	Actively working a task	Wait, or synapse interrupt if stuck
WAITING	Awaiting a permission/approval prompt	synapse approve / synapse reject
WAITING_FOR_INPUT	Task is paused asking for non-permission input (#538)	synapse reply <task_id> with the answer
RATE_LIMITED	Last task failed due to LLM provider rate limit (#561)	Wait for the provider window to reset, then re-send
DONE	Task complete; demotes to READY after ~10s	Read result, then proceed
SHUTTING_DOWN	Agent is exiting	Do not send

Stuck on a CLI dialog (not A2A WAITING)? When an agent looks idle but is blocked on its own TUI prompt (codex edit-confirmation, model picker, rate-limit dialog), use synapse send-keys <target> <keys> to write directly to the PTY without synapse jump. Example: synapse send-keys Impl a sends the codex "don't ask again" shortcut. (#695)

Killing spawned agents after completion frees ports, memory, and PTY sessions, and prevents orphaned agents from accidentally accepting future tasks.

# Preferred: one-command spawn + delegate (handles readiness wait internally)
synapse spawn gemini \
    --name Tester \
    --role "test writer" \
    --task-file /tmp/test-spec.md \
    --task-timeout 600 \
    --notify
# (do other work; receive async A2A notification when Tester finishes)
# Evaluate result, then cleanup
synapse kill Tester -f
synapse list --json                       # Verify cleanup (AI-safe)

If synapse kill fails or the agent still appears in synapse list --json, retry with -f, check the agent status/logs, and report the cleanup failure instead of leaving an orphaned agent behind.

Response Mode Guide

Choose based on whether you need the result:

Mode	Flag	Use When
Wait	--wait	You need the answer before continuing (questions, reviews)
Notify	--notify (default)	Async — you'll be notified on completion
Silent	--silent	Fire-and-forget delegation (no response needed; sender history still updates best-effort on completion)

Worker Agent Guide

When you receive a task from a manager:

On Task Receipt

1. Start work immediately (synapse reply is valid for Synapse-tracked messages with a registered reply target, including [REPLY EXPECTED] and --response; otherwise user-pasted A2A: text has no reply target, so respond with synapse send instead)
2. Check shared knowledge: synapse memory search "<task topic>"
3. Lock files before editing (skip if SYNAPSE_WORKTREE_PATH is set): synapse file-safety lock <file> $SYNAPSE_AGENT_ID

During Work

• Report progress if task takes >5 minutes: synapse send <manager> "Progress: <update>" --silent
• Report blockers immediately: synapse send <manager> "<question>" --wait
• Save findings: synapse memory save <key> "<finding>" --tags <topic>
• You can delegate subtasks too — spawn helpers (prefer different model types)
• Always clean up agents you spawn: synapse kill <name> -f

On Completion

1. Report to manager: synapse send <manager> "Done: <summary>" --silent

On Failure

1. Report details: synapse send <manager> "Failed: <error details>" --silent

Related Skills

Skill	Purpose
synapse-manager	Multi-agent orchestration workflow (delegation, monitoring, verification)
synapse-reinst	Re-inject instructions after /clear or context reset

References

For detailed information, consult these reference files:

Reference	Contents
references/commands.md	Full CLI command documentation with all options
references/api.md	A2A endpoints, readiness gate, error handling
references/examples.md	Multi-agent workflow examples and patterns
references/file-safety.md	File locking workflow and commands
references/messaging.md	Sending, replying, priorities, status states, interactive controls
references/spawning.md	Spawn lifecycle, patterns, worktree, permissions, API
references/collaboration.md	Agent naming, external agents, auth, resume, path overrides
references/features.md	Sessions, workflows, saved agents, tokens, skills, settings, Canvas