solely-responsible-agent

name	solely-responsible-agent
description	Design agents that are solely responsible for one unique concern of a project — project avatars, watch officers, wardens, sentinels — with exclusive scope, a durable state home, a mandatory ledger, an escalation path, and a handover protocol. Use when creating a per-project guardian or log-watching agent, when promoting an existing fleet agent (test-hunter, spark, spider) to true sole ownership of its concern, or when choosing where a long-lived agent's state should live (notes vs tuples vs episodes vs own tables). NOT for one-shot task agents, general multi-agent orchestration topology, skill authoring, prompt engineering, or CI pipeline design.
allowed-tools	Read,Write,Edit,Bash,Grep,Glob
argument-hint	[concern, e.g. "log watching for port-daddy"] [action: design|instantiate|audit]
metadata	{"category":"Agent & Orchestration","tags":["sole-responsibility","project-avatar","watch-officer","agent-state","accountability"],"pairs-with":[{"skill":"port-daddy-internal-dev","reason":"Instantiating an avatar in the port-daddy repo follows that repo's fleet/PR conventions"},{"skill":"multi-agent-coordination","reason":"Sole owners are nodes in a wider coordination topology; this skill designs one node deeply"},{"skill":"ostrom-commons-governance","reason":"Graduated sanctions and monitoring for violated obligations come from commons governance"}]}
io-contract	{"kind":"deliverable","produces":[{"kind":"design-doc","description":"Sole-responsibility agent specification covering exclusive scope, durable state home, mandatory ledger, and escalation/handover protocol"}]}

Solely Responsible Agent: Design One Owner Per Concern

Things nobody owns, nobody does. An error message can flood a project's message traffic for weeks with a dozen agents active — because every agent's job description says "do your task," and no agent's says "read the traffic." This skill designs agents whose entire identity is one concern of one project: the avatar, the officer of the watch, the test warden. One agent, one concern, exclusive ownership, durable state, mandatory logging, real enforcement.

Philosophy

Responsibility = scope + authority + state + ledger + enforcement. Miss any one and the responsibility is decorative:

• Scope without exclusivity → shared accountability, which disappears ("each assumes the other is watching").
• Authority withheld → the responsibility trap: blame rolls downhill, control stays up.
• No private state → the agent starts blind every run; it can detect events but never trends.
• No mandatory ledger → a silent agent is indistinguishable from a dead one.
• No enforcement → the obligation is a prompt suggestion. Prompts are suggestions; code is enforcement.

The strongest precedent is 400 years old: the ship's officer of the watch has exclusive scope (this watch period), real authority (can alter course, must wake the captain per standing orders), a mandatory append-only log (the deck log, signed, never torn out, admissible as evidence), and a handover ceremony (brief the relief, record it).

Design Process

flowchart TD
    A[Name the concern] --> B{Is exactly one agent\naccountable for it?}
    B -- "shared / fuzzy" --> A2[Split or narrow until\nRACI has ONE Accountable]
    A2 --> B
    B -- yes --> C[Grant matching authority:\ntools, escalation rights,\nbudget]
    C --> D{Pick the state home}
    D --> D1[Operator-visible findings\n→ immutable typed notes]
    D --> D2[Agent working memory\n→ episodes / own table-namespace]
    D --> D3[Live signals to other agents\n→ tuples with TTL / pheromones]
    D1 & D2 & D3 --> E[Define the reconcile loop:\ncover the gap since last\nledger entry, not a fixed window]
    E --> F[Mandate the ledger:\none entry per cycle,\nALL QUIET included]
    F --> G[Define escalation tiers:\nlog-only → issue → page]
    G --> H[Write the handover:\neach entry ends with notes\nfor the next cycle]
    H --> I{Enforcement available?}
    I -- yes --> J[Register the obligation:\nmonitor + graduated sanctions]
    I -- "not yet" --> K[Document the gap honestly;\nnext-cycle detection of a\nmissing entry is the floor]

Step details

1. Name the concern as a question. "Did anything land in this project's logs that a human would want to know?" is ownable. "Help with quality" is not. The agent's telos is the question; its ledger is the running answer.