菜单
Synchronous immediate signaling channel for inter-agent communication. Implements latest-wins single-file nudge pattern for health checks, stall detection, and urgent pings.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/Tibsfox/gsd-skill-creator --skill nudge-sync命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Use the moment a request is underspecified and you are about to act on an ASSUMED goal. Enumerate the candidate goals the request could mean, project the action or artifact each one implies, and check whether they diverge: if two or more plausible goals lead to materially different actions or artifacts, ask exactly ONE targeted clarifying question; otherwise proceed on the most-likely goal and state the assumption in one line. Clarification is an evidence-producing action, not a delay — but one question, never an interrogation. Distinct from gsd-spec-phase (GSD-phase-bound, emits a heavy SPEC.md) and intent-router (fetch strategy, not goal disambiguation). Backed by agent goal-state inference (arxiv 2606.16813v1). Triggers on acting under an assumed goal when the request admits more than one materially different reading.
Use at write time to vector/embedding memory — Grove content-addressed insertion, chroma/pgvector upserts, memory-consolidation promoting session traces to MEMORY.md, or embedding externally-ingested content. Scores each candidate record against a fixed panel of sentinel queries and quarantines any record that would become the nearest neighbor of too many unrelated queries — a hub — whether from adversarial poisoning or accidental over-generality. This is the memory-record-side sibling of skill-injection-guardian (file-side) and the write-side complement of memory-use-warrant (read-side). Quarantine, never silently drop; a human reviews. Backed by the admission-time hubness gate (arxiv 2606.19692v1). Triggers on inserting into vector memory, consolidating memory, and embedding stranger content.
Run this appropriateness check the moment you are about to integrate a retrieved long-term memory — a Grove content-addressed hit, a chroma/pgvector neighbour, a memory-consolidation digest, or a MEMORY.md line — into a response, especially anything touching private origins, Fox Companies IP, credentials, or Center Camp / consent-governed content. It answers a question intent-router never asks: not WHAT to fetch or HOW, but WHETHER a correctly-retrieved item should reach output. Relevance is not appropriateness — a perfect similarity match can still be a boundary violation. Default is FAIL-CLOSED: if in-context authorization is uncertain, the memory may inform behaviour but must not be surfaced. Backed by RBI-Eval (arxiv 2606.06055v1). Triggers on surfacing recalled sensitive memory into a response.
正在显示 SKILL.md
基于 SOC 职业分类
| name | nudge-sync |
| description | Synchronous immediate signaling channel for inter-agent communication. Implements latest-wins single-file nudge pattern for health checks, stall detection, and urgent pings. |
| type | skill |
| category | state |
| status | stable |
| origin | tibsfox |
| modified | false |
| first_seen | "2026-03-06T00:00:00.000Z" |
| first_path | .claude/skills/nudge-sync/SKILL.md |
| superseded_by | null |
Lightweight synchronous signaling for multi-agent orchestration. Each agent has a single nudge file that is overwritten on every new nudge -- latest wins, no accumulation. Agents check their nudge file on every state poll, making this the fastest communication channel in the chipset.
Nudge is the low-bandwidth, urgent signal channel in the Gastown chipset -- the SMI (System Management Interrupt) equivalent. It carries health checks from the witness, stall recovery prompts, and urgent coordination signals. Unlike mail (which accumulates), nudge is intentionally ephemeral: only the latest nudge matters.
The witness uses nudge to implement Gastown's Deacon heartbeat pattern. When an agent has hooked work but hasn't reported activity, the witness sends a nudge asking "are you working?" If the agent doesn't respond within the nudge interval, the witness escalates to the mayor via mail.
.chipset/state/nudge/{agent-id}/latest.json
Each agent has a dedicated nudge directory containing exactly one file: latest.json. This file is overwritten on every new nudge. Reading the file always returns the most recent nudge (or null if no nudge has been sent).
Example paths:
.chipset/state/nudge/polecat-alpha/latest.json
.chipset/state/nudge/mayor-a1b2c/latest.json
.chipset/state/nudge/refinery-f5g6h/latest.json
{
"from": "witness-d3e4f",
"type": "health_check",
"message": "You have hooked work (gt-abc12). Are you working on it?",
"timestamp": "2026-03-05T10:35:00Z",
"requires_response": true
}
| Field | Type |
|---|
| Required |
|---|
| Description |
|---|
from | string | yes | Sender agent ID |
type | string | yes | Nudge type (see Nudge Types below) |
message | string | yes | Human-readable nudge content |
timestamp | string | yes | ISO 8601 creation timestamp |
requires_response | boolean | yes | Whether the recipient must respond |
| Type | Sender | Purpose |
|---|---|---|
health_check | witness | Verify agent is alive and working |
stall_warning | witness | Agent has not reported activity |
priority_change | mayor | Work item priority has changed |
abort | mayor | Stop current work immediately |
sync_request | any | Request state synchronization |
The send protocol overwrites the recipient's latest.json atomically.
.chipset/state/nudge/{agent-id}/.nudge.tmplatest.json (atomic overwrite)async function sendNudge(nudge: NudgeMessage): Promise<void> {
const nudgeDir = join(stateDir, 'nudge', nudge.to);
await mkdir(nudgeDir, { recursive: true });
const filePath = join(nudgeDir, 'latest.json');
const content = serializeSorted(nudge);
const tmpPath = join(nudgeDir, '.nudge.tmp');
const fd = await open(tmpPath, 'w');
try {
await fd.writeFile(content, 'utf-8');
await fd.sync();
} finally {
await fd.close();
}
await rename(tmpPath, filePath);
}
Note: The nudge message includes a to field implicitly via the directory path. The JSON itself does not store to -- the recipient is identified by the directory.
Agents read their latest.json on every state poll cycle. If the file exists and has a newer timestamp than the last processed nudge, the agent processes it.
async function checkNudge(agentId: string): Promise<NudgeMessage | null> {
const filePath = join(stateDir, 'nudge', agentId, 'latest.json');
return readJson<NudgeMessage>(filePath);
}
When an agent reads a nudge, it decides how to respond based on type and requires_response:
async function processNudge(agentId: string, lastSeen: string): Promise<void> {
const nudge = await checkNudge(agentId);
if (!nudge) return;
if (nudge.timestamp <= lastSeen) return; // Already processed
switch (nudge.type) {
case 'health_check':
if (nudge.requires_response) {
await respondToNudge(agentId, nudge);
}
break;
case 'stall_warning':
// Log warning, update activity timestamp
await updateActivity(agentId);
if (nudge.requires_response) {
await respondToNudge(agentId, nudge);
}
break;
case 'abort':
// Stop current work, clear hook
await abortWork(agentId);
break;
case 'priority_change':
// Re-read hook for updated priority
await refreshHook(agentId);
break;
case 'sync_request':
// Synchronize state
await syncState(agentId);
break;
}
}
When requires_response is true, the agent must write a response within the nudge interval. Responses are written to the sender's nudge directory (the sender becomes the recipient).
type: "nudge_response"latest.jsonasync function respondToNudge(
agentId: string,
incoming: NudgeMessage
): Promise<void> {
const response: NudgeMessage = {
from: agentId,
type: 'nudge_response',
message: `Acknowledged. Working on hooked bead. Last activity: ${new Date().toISOString()}`,
timestamp: new Date().toISOString(),
requires_response: false,
};
// Write to the sender's nudge directory
await sendNudge({ ...response, to: incoming.from });
}
The nudge interval defines how long a sender waits for a response before escalating. Default: 60 seconds. Configurable per-agent in the chipset configuration.
If no response arrives within the interval:
health_escalation mail to the mayorThe witness implements Gastown's Deacon pattern using nudge:
[Witness] --health_check--> [Polecat]
|
(working? yes)
|
[Witness] <--nudge_response-- [Polecat]
If the polecat doesn't respond:
[Witness] --health_check--> [Polecat]
|
(no response)
|
[Witness] --health_escalation (mail)--> [Mayor]
|
(reassign work)
The witness runs the Deacon loop on a configurable interval, checking all agents with active hooks.
Nudge intentionally discards history. If two nudges arrive in rapid succession, only the second one is visible to the recipient. This is by design:
This contrasts with mail-async, which accumulates all messages.
Nudge works with the other channels in the escalation flow:
Nudge is the detection layer. Mail is the escalation layer. Hook is the assignment layer.
| Condition | Behavior |
|---|---|
| Nudge directory doesn't exist | Created automatically on first send |
Corrupt latest.json | Returns null, treated as no nudge pending |
| Concurrent nudge writes | Last writer wins (atomic rename) |
| Agent terminated | Nudge file persists but is stale (timestamp check prevents reprocessing) |
| Response timeout | Escalation via mail-async to mayor |
latest.json. No filename variationsrequires_response is true