setup-fstack
Skill18 starsChanged 53 days agoMITView on GitHub
Configure which models fstack uses per role and at what reasoning budget. Detects available models and writes a per-harness override sheet. Use for /setup-fstack, "configure fstack models", "fstack budget", or changing model choices.
In this repository75 other files it gives its agents.
- AGENTS.mdAGENTS.md
- architectSkill
- arenaSkill
- automate-meSkill
- babysitSkill
- blast-radiusSkill
- broSkill
- browseSkill
- ceo-reviewSkill
- changelog-to-postSkill
- create-verification-skillSkill
- customer-lensSkill
- deslopSkill
- engineer-modeSkill
- figure-it-outSkill
- fix-ciSkill
- fix-merge-conflictsSkill
- founder-modeSkill
- founder-voiceSkill
- fstackSkill
- geo-pageSkill
- get-pr-commentsSkill
- howSkill
- inbox-triageSkill
- interrogateSkill
- lead-replySkill
- maintain-verification-skillSkill
- make-pr-easy-to-reviewSkill
- no-commentsSkill
- office-hoursSkill
- poteto-modeSkill
- principle-attack-the-premiseSkill
- principle-boundary-disciplineSkill
- principle-build-the-leverSkill
- principle-clear-namingSkill
- principle-encode-lessons-in-structureSkill
- principle-exhaust-the-design-spaceSkill
- principle-experience-firstSkill
- principle-fix-root-causesSkill
- principle-foundational-thinkingSkill
- principle-guard-the-context-windowSkill
- principle-laziness-protocolSkill
- principle-make-operations-idempotentSkill
- principle-migrate-callers-then-delete-legacy-apisSkill
- principle-minimize-reader-loadSkill
- principle-model-the-domainSkill
- principle-never-block-on-the-humanSkill
- principle-outcome-oriented-executionSkill
- principle-prove-it-worksSkill
- principle-redesign-from-first-principlesSkill
- principle-separate-before-serializing-shared-stateSkill
- principle-sequence-verifiable-unitsSkill
- principle-subtract-before-you-addSkill
- principle-test-behavior-not-implementationSkill
- principle-type-system-disciplineSkill
- recallSkill
- reflectSkill
- retroSkill
- show-me-your-workSkill
- social-postSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 9
Write a per-harness override sheet so /engineer-mode delegations use the models you actually have.
- Very specific: line 27
If the override file for this harness already exists, read it and treat its # budget line and its role values as current.
- Distinctive: line 31
(a) Ask for a budget: Prefer a structured question over free text.
- Very specific: line 46
Every real slug written must be in the detected set.
- Very specific: line 82
Claude Code / Codex: use the same role rows and the same # budget line.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Setup fstack
Write a per-harness override sheet so /engineer-mode delegations use the models you actually have. Repo-root models.json is the default map. The override sheet wins when present.
| Harness | Override file | How it loads |
| Cursor | ~/.cursor/rules/fstack-models.mdc | alwaysApply: true rule |
| Claude Code | ~/.claude/fstack-models.md | @~/.claude/fstack-models.md in CLAUDE.md |
| Codex | ~/.codex/fstack-models.md | paste into ~/.codex/AGENTS.md |
On Codex, slugs are Codex models (for example gpt-5.5), not claude-*. Detect them from ~/.codex/config.toml plus what the user confirms. See codex-tools.md.
Steps
1. Detect available models
Enumerate the model slugs you can pass to a subagent in this session. That is the dependable source. Never write a real slug you have not confirmed is available. inherit-parent and auto are always valid. Both mean the role runs on the parent session's model (omit model on the subagent call).
2. Load current state
If the override file for this harness already exists, read it and treat its # budget line and its role values as current. Otherwise start from models.json for this harness, falling back to universal.
3. Budget, map, and confirm
(a) Ask for a budget. Prefer a structured question over free text. Offer these four options with these exact labels, and name the current budget when the sheet records one.
unlimited - keep maxlarge - xhigh reasoningmedium - high reasoningsmall - medium reasoning
(b) Apply it. Build the working table from the loaded state. On a re-run keep any role you changed by family, list, or alias (inherit-parent, auto). unlimited leaves every effort as in that table. large, medium, and small set the effort token of every real slug, panel entries included, to xhigh, high, or medium. The effort token is the last token, or the one before a trailing fast, on the ladder max > xhigh > high > medium > low. If the result is not a detected slug, use the same family's detected slug with the highest effort at or below the target, else mark the role as needing a choice. inherit-parent and auto do not change. Do not invent vendor slugs. Rewrite effort tokens on whatever models.json or the existing override already has.
(c) Show the roles and confirm. Show every role with its model. Mark any real slug not in the detected set as needing a choice. Ask whether to accept as-is or change specific roles. Prefer a structured question over free text.
Panel roles (how critics, arena runners, architect runners, interrogate reviewers) are lists. One subagent runs per entry. arena cross-judge pool is also a list. Arena picks one value whose model family differs from the parent when possible. swarm workers is the default worker model unless a race assigns another model per arm.
4. Validate
Every real slug written must be in the detected set. inherit-parent and auto always pass. An override pointing at a model the user cannot use breaks every delegation that reads it.
5. Write the override
Overwrite the whole file so re-runs stay idempotent. Include a # budget line with the chosen label and its target effort.
Cursor (~/.cursor/rules/fstack-models.mdc):
---
description: fstack per-role model choices (overrides models.json)
alwaysApply: true
---
# fstack model configuration. One line per role. Delete a line to fall back to models.json.
# inherit-parent or auto: the role runs on the parent chat model (omit Task `model`).
# budget: unlimited (max)
feature, refactoring: composer-2.5
bug-fix: gpt-5.6-sol-high
perf-issue: gpt-5.6-sol-high
hillclimb: gpt-5.6-sol-high
judgment and prose: claude-opus-5-thinking-high
hardest tasks: gpt-5.6-sol-high
how explorer: composer-2.5-fast
how explainer: composer-2.5
how critics: composer-2.5, gpt-5.6-sol-high, cursor-grok-4.6-high
why investigators: composer-2.5-fast
why synthesizer: composer-2.5
reflect tooling: composer-2.5-fast
reflect judgment, divergent, synthesizer: gpt-5.6-sol-high
arena runners: composer-2.5, gpt-5.6-sol-high, cursor-grok-4.6-high
arena cross-judge pool: composer-2.5, gpt-5.6-sol-high, cursor-grok-4.6-high
swarm workers: composer-2.5-fast
architect runners: composer-2.5, gpt-5.6-sol-high, cursor-grok-4.6-high
interrogate reviewers: composer-2.5, gpt-5.6-sol-high, cursor-grok-4.6-highClaude Code / Codex use the same role rows and the same # budget line. Change only the slugs and the file path.
6. Wire it in
Cursor: the .mdc rule applies to new sessions automatically.
Claude Code: if ~/.claude/CLAUDE.md does not already include ~/.claude/fstack-models.md, append @~/.claude/fstack-models.md.
Codex: append the sheet's contents to ~/.codex/AGENTS.md.
7. Confirm
Tell the user where the override was written and that re-running this skill updates it.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

