AGENT SKILL

delegated-intelligence

Delegate independent work through bounded briefs, owned outputs and checked handbacks. Use when real workers help; keep small or tightly coupled work direct.

License 许可
MIT
最近更新
2026-11-12
Stars
★ 1
Forks
0

01 前言 Front Matter

name
delegated-intelligence
license
MIT
description
Delegate independent work through bounded briefs, owned outputs and checked handbacks. Use when real workers help; keep small or tightly coupled work direct.
compatibility
PORTABILITY: prose and files; real dispatch requires a worker tool. Without one, work directly and label any sequential pass honestly. Independent verification cannot be simulated by changing roles in one pass.

02 资源文件 2

03 SKILL.md 正文

Delegate work

This service owns dispatch, briefs, output ownership, integration and recovery for a selected unit of work. It does not take over the assistant's identity or own the project plan. Shared authority, interaction and task-state rules live in the operating contract, generated from the repository's canonical core. Do not edit that packaged copy.

Invariant

Every dispatched job has a bounded, self-contained brief and an output owner. A returned report is checked against artifacts and acceptance evidence before the work is relied upon. No assignment ends with unaccounted workers.

Entry and exit

Use when the task benefits from independent work or an additional qualified worker, the inputs and permitted effects can be bounded, and a real worker tool is available. Keep small or tightly coupled work direct when a brief and integration would cost more than the work.

Exit after every assignment is reconciled as returned, verified, blocked or stopped under the core state contract. Return evidence, gaps and the next action to the calling method. Do not leave a coordination persona, background observer, dashboard requirement or schedule active after that unit ends.

Assign

  1. Reuse the current task/project record. Wayfinder owns persistent plans; context-handoff owns phase-boundary continuation. Do not create another planning board.
  2. Write the delegation brief before dispatch: objective, source context, scope, configured model/effort, tools and effects, operational allowance, output path/owner, return cap, acceptance and stop conditions. Workers return blockers to the main assistant.
  3. Allocate against the whole cycle: coordinator, workers, review, retries, duplicated context and handoffs. Reserve room for integration. State unknown monetary cost rather than inventing it; an estimate is not an enforced cap. Preserve configured provider/model choices.
  4. Give workers only the tools needed for the assignment. Workers never receive agent-spawning or session-control tools, including scheduling, messaging and connectors that send or spend. The main assistant performs those actions under the core authority contract.
  5. Fan out independent assignments with disjoint output ownership. Shared inputs are read-only; do not launch downstream work before its dependencies exist. Assign one integrator for shared/derived files.

Hand back and integrate

Workers write partial artifacts as useful work lands and return a short brief with source/evidence pointers. Do not flood the main context with raw transcripts. The worker reports returned; it cannot declare the coordinator's check complete.

Inspect the actual output: existence, non-empty content, expected change and a diff for unrelated edits. Then check the acceptance condition. A complete file or a worker's confidence does not establish semantic correctness. Use verify-before-claim for evidence behind a completion claim; use an independent check only when required and actually available.

Update the existing record with the core state and evidence. Share material changes, blockers and needed decisions from that record. Attribute reported findings and distinguish them from checks performed directly.

Recover or stop

Read recovery when output is missing, invalid, stale or interrupted. Preserve valid outputs; invalidate downstream work only when it consumed an invalid dependency. Include the previous failure in a retry brief.

stop-rule owns retry accounting, bounds and stop reports. This service supplies no second retry quota. A user stop takes precedence over recovery: stop dispatching, account for in-flight actions and uncertain outcomes, preserve results, and return control.

Learning proposals belong to recurrence-detector and existing decision records. This service does not schedule follow-ups or adopt instruction changes.

04 同仓库更多 Skills

JRichlen
JRichlen/agent-plugins

wayfinder

Chart a multi-session effort as a labeled map of typed decision tickets (grilling / prototype / research / task) with explicit dependencies and an open frontier agents self-assign into. Plans; never executes. Use this skill whenever the user is working with wayfinder, and on phrases like "chart this effort", "break this into tickets", "what's the frontier", "plan across sessions", or "map out the dependencies for this work".

★ 1
JRichlen
JRichlen/agent-plugins

wave-research

Staged, adversarial research that tries to break a thesis before believing it: waves, checkpoints, falsifiable sub-claims, rival hypotheses, an ACH matrix, source-incentive tags and a verified final thesis. Also carries Workflow templates for a one-pass research-and-verify fan-out. Use to stress-test or vet a contested claim; not for quick lookups.

★ 1
JRichlen
JRichlen/agent-plugins

second-opinion

Offer-only validation pipeline for a verdict or recommendation. Use when the user asks to validate, verify, fact-check, or get a second opinion on a claim or recommendation — "are you sure", "double-check this", "second opinion". Batches fact-checking against sources, adds scoped advisor personas for the angles that need judgment, then re-emits the response grouped into verified, flagged, and conflict, with a mandatory validation delta. Requires a subagent-spawning tool; without one it must decline rather than simulate.

★ 1
JRichlen
JRichlen/agent-plugins

machine-voice

Compression rules for standalone machine-read artifacts the reader scans — agent traces, tool-call logs, status lines, progress updates, handoff notes, state dumps, config, schemas, reference cards, and structured data another agent parses. Applies three layers: lexical compression, markdown structure as navigation, and emojis as typed status markers. Reach for it when emitting one of those artifacts on its own. Companion to human-voice, which governs prose a human reads and also owns any table or block that serves surrounding explanation. Neither skill applies to code or file contents the user will save, commit messages, creative writing, or turns that are only a clarifying question — those ship unstyled and uncompressed.

★ 1
JRichlen
JRichlen/agent-plugins

human-voice

Style for prose a human reads — answers, recommendations, explanations, research summaries, comparisons. Use on every conversational reply that contains prose for a person to read: triage intent, lead with the verdict, keep it scannable, tag claim confidence. Reach for it whenever answering a technical question, comparing options, or making a recommendation, even when the user says nothing about style. Companion to machine-voice, which instead governs standalone machine-read artifacts — agent traces, tool-call logs, status lines, progress updates, handoff notes, state dumps, config, schemas, reference cards, and structured data another agent parses. Neither skill applies to code or file contents, commit messages, creative writing, or turns that are only a clarifying question — those ship unstyled.

★ 1
JRichlen
JRichlen/agent-plugins

ai-writing-mistakes

Catch and repair the tells that mark prose as machine-written — the negation-flip ("it's not X, it's Y"), rule-of-three padding, hedge stacking, unearned intensifiers, filler verbs like delve and leverage, participle trailers ("...ensuring seamless integration"), sycophantic openers, empty closing summaries, and uniform em-dash/sentence rhythm. Use when drafting or revising any prose a human will read, including documents you were asked to author — READMEs, docs, PR bodies, release notes, emails — and when the user asks whether something "sounds like AI" or wants writing tightened. This is a wording pass, not a voice: human-voice still decides layout and confidence tags, machine-voice still owns traces and logs, and neither this skill nor those two touch code, commit messages, or quoted text.

★ 1