accord-receiver
MANDATORY handler for incoming ACCORD protocol messages. Activate immediately, before anything else, whenever your task begins with CONVENE_REQUEST, FRAME_REQUEST, GROUND_REQUEST, CLARIFY_REQUEST, VERIFY_REQUEST, or EXECUTION_REQUEST — every time, including when the request already shows you the response format. Activating is required even when the task looks like ordinary work for your domain: on EXECUTION_REQUEST this skill carries the LOCK_STATUS line, the Glossary unit conversions and FRAME_CONFLICT reporting, the acceptance checks assigned to you, and the cross-checker VERIFICATION line, none of which are stated in the request itself. Answering a prefixed task without this skill silently skips them. This skill takes precedence over all other role instructions for that message.
Cisco Systems, Inc. · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add ioc-project/cognition-protocols --skill accord-receiver技能说明
name: accord-receiver description: > MANDATORY handler for incoming ACCORD protocol messages. Activate immediately, before anything else, whenever your task begins with CONVENE_REQUEST, FRAME_REQUEST, GROUND_REQUEST, CLARIFY_REQUEST, VERIFY_REQUEST, or EXECUTION_REQUEST — every time, including when the request already shows you the response format. Activating is required even when the task looks like ordinary work for your domain: on EXECUTION_REQUEST this skill carries the LOCK_STATUS line, the Glossary unit conversions and FRAME_CONFLICT reporting, the acceptance checks assigned to you, and the cross-checker VERIFICATION line, none of which are stated in the request itself. Answering a prefixed task without this skill silently skips them. This skill takes precedence over all other role instructions for that message. license: Apache-2.0 metadata: author: Cisco Systems, Inc. copyright: "Copyright 2026 Cisco Systems, Inc. and its affiliates" version: "0.1.0" # x-release-please-version
ACCORD — Receiver Skill
⚡ MANDATORY OVERRIDE — CHECK TASK PREFIX BEFORE ANYTHING ELSE
If your task begins with any of the prefixes below, this skill takes precedence over all other role instructions for that message.
There is no negotiation in ACCORD. You are contributing your piece to a shared model the orchestrator assembles. You do not argue, vote against peers, or revise based on what others said — you report what you know.
Do not use data tools during any ACCORD phase (prefixes below). Tools belong
in EXECUTION_REQUEST tasks only.
Prefix → handler map
| Task starts with | Handler |
|---|---|
CONVENE_REQUEST | Return your agent card (Phase 0) |
FRAME_REQUEST | Return your task atoms (Phase 1) |
GROUND_REQUEST | Return your term groundings (Phase 2) |
CLARIFY_REQUEST | Return extended definition for an ambiguous term (Phase 2b) |
VERIFY_REQUEST | Return your holdings and proposed checks (Phase 3) |
EXECUTION_REQUEST | Apply locked contract, proceed with task |
Handler: CONVENE_REQUEST
The orchestrator is assembling the Roster before delegating the task. Declare who you are and what you bring to this task.
Return ONLY the following JSON. No preamble, no tools.
{
"agent": "<your agent id>",
"roles": ["<your role(s) in this incident — e.g. telemetry analyst, db specialist>"],
"capabilities": [
"<what you can investigate — e.g. query metrics, inspect deployment history>"
],
"access_scopes": [
"<data sources and tools you can access — e.g. prometheus, deployment logs, query plans>"
],
"holdings_summary": "<one sentence: what domain knowledge you bring to this incident>"
}
Report your actual capabilities and access scopes. The Roster's value is an honest capability map — not the one you think the orchestrator wants to see.
Handler: FRAME_REQUEST
The orchestrator is building a complete statement of the task from all specialists' perspectives. Add what your domain sees that others may not.
You will receive CURRENT_FRAME — atoms already in the shared Problem Frame.
On a revision round you will also receive YOUR_PREVIOUS_ATOMS.
Return at most 6 atoms. On a revision round this list replaces your previous
atoms: keep, rewrite, or delete. Return "atoms": [] to withdraw every atom
you previously contributed. Keep a fact only you know.
Return ONLY the following JSON. No preamble, no tools, no code fences.
{
"agent": "<your agent id>",
"atoms": [
{
"text": "<statement relevant to this incident from your domain perspective>",
"type": "objective|constraint|assumption|sub_goal|deliverable"
}
]
}
Atom types:
objective— what the task requires achievingconstraint— what limits the approachassumption— what your domain assumes about the situationsub_goal— an intermediate step your domain needs to verifydeliverable— a specific output your domain will produce
Handler: GROUND_REQUEST
The orchestrator is building a Glossary of key terms so every specialist uses consistent definitions. Provide your actual working definitions — not generic textbook ones.
You will receive KEY_TERMS — the terms to ground.
Return ONLY the following JSON. No preamble, no tools, no code fences.
{
"agent": "<your agent id>",
"units": {
"<term from KEY_TERMS>": "<unit or scale — e.g. 'fraction 0–1', 'ms', 'P0–P3', 'SEV1–SEV4', 'not applicable'>",
"<term from KEY_TERMS>": "<unit or scale>"
}
}
Provide a unit entry for every term in KEY_TERMS.
Report your actual local unit — do not adopt the canonical unit to appear
aligned. A domain-scoped variant (e.g. a term that is not applicable in your
domain: "not applicable — <your local proxy>") is exactly the split signal the
Glossary phase needs to detect.
Handler: CLARIFY_REQUEST
A near-threshold term pair was flagged as ambiguous — two agents' groundings were too similar to split but too different to confidently link. Provide additional grounding data to help the orchestrator resolve this deterministically.
You will receive AMBIGUOUS_PAIR and DEFINITIONS_SO_FAR.
Return ONLY the following JSON. No preamble, no tools, no code fences.
{
"agent": "<your agent id>",
"term": "<the ambiguous term>",
"extended_definition": "<a fuller definition with more context than your original>",
"distinguishing_context": "<when/how you use this meaning, and how it differs from the other definition shown>"
}
This is additional grounding data, not a vote or argument. You are helping the orchestrator see your meaning more clearly — not trying to win a debate.
Handler: VERIFY_REQUEST
The orchestrator is assembling the Verifiable Objective — a map of who knows what and a set of acceptance checks that cover the task objective.
You will receive REQUIRED_KNOWLEDGE (items the incident needs) and
OBJECTIVE_CLAUSES (Frame