Skills and setup for Claude and Codex.
Choose what to install for your coding agent. Start with Workflow for planning, changes, review, and writing; these commands install it.
Claude Code
claude plugin marketplace add filipgutica/filip-stack
claude plugin install workflow@filip-stackCodex
codex plugin marketplace add filipgutica/filip-stack
codex plugin add workflow@filip-stack- Hosts
- Claude Code and Codex
- Skills
- 5 Workflow, 19 standalone
- Source
- filipgutica/filip-stack
Choose your setup
- Workflow
Install it to get all 5 Workflow skills.
- Standalone skills
Choose individual tools for PRs, Jira, Vue, and UX. They install separately from Workflow.
- Field Guide
Add this optional plugin to keep preferences and lessons across sessions.
- filip-stack CLI
Run it to apply shared guidance, agent profiles, output styles, and the design system on each machine.
- Design system
Shared visual rules for pages, mockups, and explainers. The CLI installs them to
~/design-system.
How Workflow works
Workflow is 5 skills that install together. You describe a task in Claude or Codex in your own words, without naming a skill, though you can. For each request, the host compares it with each skill's name and description and picks the skill that fits.
The picked skill reads its own SKILL.md. It reads a longer guide from a references folder only when the task needs it.
Example: one feature, four requests
You type
“Write a spec for rate limiting on the public API.”
The host picks
planning. It writes the spec from what the repository shows and changes no code. A public API is a public contract, so a separate reviewer critiques the spec before you get it.Extra guides it reads
spec.md: the format for a specificationtechnical-prose.mdfrom technical-writing: the writing rules for the spec's text
You type
“Implement the spec and open a PR.”
The host picks
engineering. It makes the smallest change that meets the spec and reruns the affected tests and checks. A public API change is risky, so a separate reviewer checks it before you get the PR. If none is available, it says so.Extra guides it reads
testing-and-debugging.md: only if it is unclear how to reproduce or test the changedelegation-and-review.md: because a public API change needs a separate reviewercommit-and-pr-hygiene.md: before it commits or opens the PR
You type
“Review the PR before I merge it.”
The host picks
review. It reports only the findings it can back with evidence and changes nothing. If the tests changed a lot, it also calls test-audit.Extra guides it reads
delegation-and-review.mdfrom engineering: to tell you whether this review counts as independent, which needs a reviewer that did not write the change
You type
“Rewrite the PR description so API users can follow it.”
The host picks
technical-writing. It rewrites the description from the branch and its checks.Extra guides it reads
pull-requests.md: the format for a PR descriptiontechnical-prose.md: the writing rules for any technical text
The 5 Workflow skills
engineeringMakes an authorized code change and verifies the requested behavior with tests and checks.
planningDevelops a direction, specification, PRD, ADR, implementation plan, or tickets from repository evidence.
reviewReviews one exact change or document and reports only the findings it can back with evidence.
test-auditAudits test value and identifies duplicate coverage, implementation-coupled tests, and test-only production code.
technical-writingWrites or revises technical documents so readers can act on them without guessing.
Some standalone skills fit before or after this flow: grill-me stress-tests a plan before the spec, and sol-loop and merge-council run final reviews before merge. They install separately and run only when asked: by you, or for merge-council, also by your repository's guidance. See Standalone skills.
Standalone skills
Replace <name> with a skill below. Run the same command again to update it.
npx skills add filipgutica/filip-stack -g -a claude-code codex -s <name> -yagent-ready-environmentsAudits or sets up repositories and agent guidance so agents can verify their work.
Picked by requestbabysit-prMonitors a PR through CI and automated review, fixing verified problems within its scope. Adapted from tstack.
Picked by requestcatch-upRecaps a thread's goal, progress, and next step in four lines when you return to it.
Manual · /catch-upcheckpointCaptures current work, evidence, and next actions for a pause or handoff.
Manual · /checkpointcli-tool-templateCreates or aligns Filip's CLI tools with shared repository, release, site, and Homebrew conventions.
Picked by requestcouncilGathers independent model opinions and synthesizes advice on a question or artifact. Adapted from tstack.
Picked by requestfleet-syncSyncs a shared filip-stack baseline across linked T3 environments with per-machine exceptions.
Picked by requestfull-sendFinishes a change or PR through review and CI, then merges or returns a verdict. Needs
Manual · /full-sendsol-loopandbabysit-pr. Adapted from tstack.full-stopFinishes a change or PR through review and CI and leaves merging to you. Needs
Manual · /full-stopfull-send,sol-loop, andbabysit-pr. Adapted from tstack.grill-meQuestions a proposal one decision at a time to expose assumptions and settle consequential choices. Adapted from Matt Pocock's skills.
Manual · /grill-mehill-climbImproves a working change or named goal in bounded rounds judged by blind comparisons, a score, or both. Needs
Manual · /hill-climbcouncil.interface-designDesigns, builds, reviews, or checks interfaces for task completion and accessibility. Adapted from frontend-design.
Picked by requestjira-work-linksCreates or checks Jira relationships and verifies the PRs and commits behind an issue.
Picked by requestmerge-councilRuns a council on a PR, fixes agreed findings, and returns a verdict tied to the reviewed commits. Needs
Picked by requestcouncil.pr-review-churnAnalyzes recurring human review objections on GitHub PRs and proposes guidance changes.
Picked by requestsimplify-passSimplifies selected code after three independent reviews, preserving its observable behavior.
Manual · /simplify-passsol-loopChecks a change through GPT Sol review rounds and resolves findings within scope. Adapted from tstack.
Picked by requestvue-best-practicesGuides Vue component, composable, and test work with explicit state ownership and component contracts.
Picked by requestwalkthroughExplains a change in parts and waits for your acceptance of each part.
Manual · /walkthrough
Field Guide
Field Guide stores preferences, corrections, and recurring lessons under ~/.field-guide. Its hooks add repository guidance at session start and matching records on your first prompt. Current instructions and live code take priority.
Install field-guide@filip-stack through the same marketplace. Workflow does not depend on it. How Field Guide works.
Guides
- Agent-ready environments
A lesson on setting up repositories so coding agents can check their own work, with a worked audit of this repository.