decision-council
Five advisors argue, review each other blind, a chairman decides, then a pre-mortem and a log entry.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/mavrin-xv--decision-council)HTML
<a href="/listings/mavrin-xv--decision-council"><img src="/api/listings/mavrin-xv--decision-council/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
/decide Northwind Roasters (2 people, 410 subscribers) wants to raise the standard plan from $12 to $18 for new customers. Margin is 9% after shipping went from $3.10 to $4.40. Competitors charge $15 and $17.
The router sizes it (high stakes, four moving parts), checks the log for killed ideas, runs the board, then a pre-mortem. Sample of the closing block from the bundled example:
DECISION: Do differently. New customers get 3 bags every 6 weeks for $16.
Existing subscribers keep $12 until November 1.
WHY: Margin is low because shipping per box rose, not because price is low.
MAIN RISK: Trial-to-paid (24% today) falls and there is no ad budget to refill it.
FIRST STEP: Email 10 current subscribers this week; quote a second courier.
KILL IF: Trial-to-paid below 18% for 2 weeks running after May 1.
SPLIT: Executor wanted a $16 vs $18 split test; too little traffic to settle it.
CONFIDENCE: 7/10The whole run, with every script output, is in `examples/northwind-price-change/`. The business and its numbers are invented.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 15 lines
- {
- "name": "decision-council",
- "plugins": [
- {
- "name": "decision-council",
- "source": "./",
- "description": "One trusted verdict on a business decision: a board of five advisors, an eight-specialist marketing council, blind peer review, chairman synthesis, pre-mortem, persona stress test, evidence-based reality check and a decision log that blocks repeats of killed ideas.",
- "version": "0.1.0"
- }
- ],
- "owner": {
- "name": "Artem Mavrin",
- "url": "https://github.com/mavrin-xv"
- }
- }
README.md · README · 153 lines
- 
- # decision-council
- **One verdict you can act on for a business decision: five advisors argue, review each other blind, and a chairman decides, then the plan gets a pre-mortem, a stress test and a log entry.**
- Asking one chat "should I do this?" gets you a polite yes. Asking five gets you five opinions and no decision.
- [](LICENSE)
- 
- 
- 
- ```bash
- git clone https://github.com/mavrin-xv/decision-council && cd decision-council
- ./install.sh # or: ./install.sh --project
- ```
- <!-- install-notes:start -->
- Other ways in: as a Claude Code plugin, run `/plugin marketplace add mavrin-xv/decision-council` and then `/plugin install decision-council@decision-council`. For Codex or another agent that reads `SKILL.md`, copy the folders under `skills/` into its skills directory. On Windows, run `install.sh` from Git Bash or WSL and type `py -3` where these docs say `python3`.
- <!-- install-notes:end -->
- Then in Claude Code: `/decide should we raise our subscription from $12 to $18?`
- ---
- ## What it solves
- | Pain | What the pack does |
- |---|---|
- | The model agrees with whatever you proposed | Five advisors with opposed frames answer independently, before seeing each other |
- | Long debates, no decision | Blind peer review with shuffled labels, then one chairman verdict with a kill criterion and a first step |
- | Marketing plans judged by one lens | Eight specialists (positioning, offer, funnel, channels, copy, brand, data, community) on the same brief |
- | Plans that fail for reasons everyone saw coming | Pre-mortem: "it is six months later and this failed, why?" with scored risks and tripwires |
- | Pages and offers shipped with holes | Six personas read it as buyers, beginners, competitors; findings merged into SHIP / REVISE / REWORK |
- | "It's done" with no proof | Reality check that defaults to NEEDS WORK until every must-have claim has evidence |
- | The same dead idea comes back every quarter | Decision log that flags proposals matching a killed decision and lists decisions due for review |
- | Running a 17-call council on a small question | A router that sizes the stakes and picks the cheapest method that fits |
- ## What's inside
- | Skill | What it does | Say |
- |---|---|---|
- | `decide` | Sizes stakes, routes to the right methods, ends with one verdict | "help me decide", "should we", "is this a good idea" |
- | `board-of-advisors` | Contrarian, First Principles, Expansionist, Outsider, Executor; blind review; chairman | "council this", "run the board", "war room this" |
- | `marketing-council` | Eight marketing specialists, blind review, chairman with first channel and first test | "marketing council", "which channel", "how do we launch this" |
- | `pre-mortem` | Plan mode: failure story, risk register, tripwires. Claim mode: doubt loop with a fresh reviewer | "pre-mortem this", "what could go wrong", "am I sure about this" |
- | `stress-test` | Six personas attack an offer, page, brief or flow; merged findings and a ship verdict | "stress test this", "find the weak spots", "before we ship" |
- | `reality-check` | Evidence gate for "done", "ready", "works" | "is it really ready", "reality check", "verify before launch" |
- | `decision-log` | Plain-markdown log with kill criteria, review dates, and a killed-idea guard | "log this decision", "what did we decide about", "what is due" |
- Each skill folder has `SKILL.md`, `references/` for the long parts, and one helper script. Slash commands live in `commands/`, blank inputs in `templates/`.
- ## Quick start
- ```text
- /decide Northwind Roasters (2 people, 410 subscribers) wants to raise the
- standard plan from $12 to $18 for new customers. Margin is 9% after
- shipping went from $3.10 to $4.40. Competitors charge $15 and $17.
- ```
- The router sizes it (high stakes, four moving parts), checks the log for killed ideas, runs the board, then a pre-mortem. Sample of the closing block from the bundled example:
- ```text
- DECISION: Do differently. New customers get 3 bags every 6 weeks for $16.
- Existing subscribers keep $12 until November 1.
- WHY: Margin is low because shipping per box rose, not because price is low.
- MAIN RISK: Trial-to-paid (24% today) falls and there is no ad budget to refill it.
- FIRST STEP: Email 10 current subscribers this week; quote a second courier.
- KILL IF: Trial-to-paid below 18% for 2 weeks running after May 1.
- SPLIT: Executor wanted a $16 vs $18 split test; too little traffic to settle it.
- CONFIDENCE: 7/10
- ```
- The whole run, with every script output, is in `examples/northwind-price-change/`. The business and its numbers are invented.
- ## How it works
- ```text
- your decision
- |
- [decide] sizes stakes ---- low ----> doubt loop on one claim -> decide
- |
- decision-log check (killed before?) --- yes ---> ask: reopen?
- |
- +----------+-----------+
- | |
- board-of-advisors marketing-council
- 5 frames, parallel 8 specialists, parallel
- | |
- council.py shuffle (labels A..E, roles hidden)
- |
- blind reviewers: strongest, blind spot, what all missed, ranking
- |
- council.py tally (where the room split)
- |
- chairman verdict: DECISION / WHY / MAIN RISK / FIRST STEP / KILL IF / CONFIDENCE
- |
- high stakes? -> pre-mortem (risk_register.py) -> doubt loop on the riskiest claim
- |
- artifact ready? -> stress-test (merge_findings.py) -> reality-check (evidence_check.py)
- |
- decision-log add (with review date) ... decision-log due (weeks later)
- ```
- Three rules carry most of the weight. Advisors never see each other in round one, so the first answer does not anchor the rest. Reviewers see answers as A to E in random order, so they judge the argument and not the role. The chairman weighs arguments, not votes, and must name the split and what nobody covered.
- Evidence labels (FACT, INTERNAL, CONVERGENCE, HYPOTHESIS) cap the confidence score: a verdict built on guesses cannot score high.
- ## Configure
- | Setting | Where | Default |
- |---|---|---|
- | Decision log folder | `DECISION_LOG_DIR` env var or `--dir` | `./decisions` |
- | Advisor frames | `skills/board-of-advisors/references/advisors.md` | five frames above |
- | Marketing specialists | `skills/marketing-council/references/specialists.md` | eight lenses |
- | Personas | `skills/stress-test/references/personas.md` | six, with domain swaps |
- | Risk bands | `skills/pre-mortem/scripts/risk_register.py` | critical from score 15 |
- Cross-model review is optional: pipe the reviewer packet to a second model's CLI and paste its JSON back. No API keys are needed by any script.
- ## Honest limits
- - A council costs model calls: about 11 for the board and 17 for the marketing council (counted from the procedure, not measured). The router exists so you do not pay that on small questions.
- - All advisors are usually the same model. Frames reduce agreement bias; they do not remove it. Use the cross-model option for decisions you cannot undo.
- - The verdict is as good as the numbers in the brief. The skills ask for sources, but cannot check your spreadsheet.
- - `decision_log.py check` matches keywords, so it misses paraphrases. The skill tells the agent to also read the brief and judge meaning.
- - The bundled example's model outputs were written by hand to show the format. Script outputs in it are real runs.
- - Tested with Claude Code. Other agents that read SKILL.md files should work; not tested.
- ## Credits
- - Based on / inspired by https://github.com/karpathy/llm-council (no license file found; idea only, no code copied): parallel first opinions, anonymized peer ranking, chairman synthesis.
- - Inspired by the five-advisor board popularised by Ole Lehmann and its open versions such as https://github.com/evisoft/council (no license file found; idea only): the five frames, random relabelling, "what did all of them miss".
- - Doubt loop based on https://github.com/addyosmani/agent-skills (MIT): fresh-context reviewer that never sees the claim, three-loop cap.
- - Reality check inspired by the Reality Checker agent in https://github.com/msitarzewski/agency-agents (MIT): default to NEEDS WORK, evidence over assertions.
- - Pre-mortem from Gary Klein, "Performing a Project Premortem", Harvard Business Review, September 2007, https://hbr.org/2007/09/performing-a-project-premortem.
- - Multi-agent debate background: Du et al., "Improving Factuality and Reasoning in Language Models through Multiagent Debate", https://arxiv.org/abs/2305.14325.
- ## Contributing
- See `CONTRIBUTING.md`. Coding agents should read `AGENTS.md` first.
- <!-- author:start -->
- ## Author
- Built and maintained by [Artem Mavrin](https://github.com/mavrin-xv). This repo is one of a set of agent skill packs for marketing and operations work. Each one installs on its own; the full list is in [agent-skill-packs](https://github.com/mavrin-xv/agent-skill-packs). If a pack saved you an afternoon, a star helps other people find it.
- <!-- author:end -->
- ## License
- MIT, see `LICENSE`.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.