Fibo Skills
When you drive an Agent CLI (Claude Code, Codex, or any Agent CLI that supports Skills) inside FiboCode's built-in terminal, Fibo Skills teach that Agent FiboCode's spec-driven workflow. They are optional — the terminal works without them — but installing them gives the Agent the same brainstorm → spec → plan → execute → sync loop that FiboCode itself follows, so its output lands in the .fibo/ document system instead of drifting away from your codebase.
Skills are packaged instructions your agent loads on demand. Fibo Skills are open source and installed into your agent, not into FiboCode.
Skills repository: github.com/Alenkz12/FiboSkills
The Skill Set
This is one workflow, not a pile of unrelated commands. The first five skills hand off from stage to stage; the last two stand on their own:
| Skill | Stage | What it's for |
|---|---|---|
fibo-brainstorming | Get it straight | Turns a vague intent into a signed-off AC list |
fibo-writing-plan | Decide the approach | Produces design.md and plan.md from the spec |
fibo-executing-plan | Write the code | Runs the plan's tasks one at a time, in order |
fibo-recording-diff | Leave a record | Turns the real diff into a readable, hunk-by-hunk diff.md |
fibo-conventions | Hold the line | One rule set for naming, comments, docs, and commit sync |
fibo-code-analysis | Read the code | Produces the tiered file → structure → folder analysis mirror |
fibo-skills-reviewer | Self-maintenance | Audits the wording of the skill files themselves |
fibo-brainstorming — pin the requirement down first
Invoked before any change to externally visible behavior. It asks one question at a time (with multiple choice whenever possible), narrowing "I want a feature" into a list of acceptance criteria in EARS syntax, and writes spec.md only after you explicitly sign off. It writes no production code, runs no tests, commits nothing — its only job is to remove ambiguity.
fibo-writing-plan — turn the spec into a design and tasks
Invoked once the spec is signed. It reads spec.md and produces, in the same directory, design.md (the HOW, with every design decision tagged ↔ AC) and plan.md (a task list where each task names its ACs, its files, and how it will be verified). If AC coverage is below 100%, or a task references an AC that doesn't exist, it refuses to write to disk.
fibo-executing-plan — execute the plan strictly serially
Runs plan.md's tasks one by one, never concurrently, so parallel tasks can't collide on the same files. Each task may only touch the files on its own list. On failure it stops and asks you (skip / abort / retry) rather than rolling back on its own. When everything is done it prints an implementation report and moves straight on to the diff record and doc sync.
fibo-recording-diff — make the change readable
Invoked after implementation. It turns a real diff into hunk-level explanations, orders the changed files into a coherent reading sequence, attaches a code anchor and — where useful — a Mermaid logic view to each hunk, and explains why each hunk exists and how it connects to the surrounding logic. It changes neither code nor AC.
fibo-conventions — one voice for docs and comments
Applies whenever you write Markdown, add code comments, or create anything under .fibo/docs/: Mermaid usage, the code / text / graph tag formats, path rules, naming rules, comment-code synchronization, and the pre-commit batch doc sync that tracks its baseline in .fibo/index.json.
fibo-code-analysis — structured analysis output
Used to generate the structured analysis under .fibo/analysis, and to understand or extend FiboCode's own analysis pipeline: prompt selection, concurrency, disk layout, and per-language prompts. It and the in-app Code Analysis are two entrances to the same way of understanding code.
fibo-skills-reviewer — a checkup for the skills themselves
Audits nothing but the wording of skill files: logical contradictions, logical gaps, unclear references, redundancy drift, leaked maintenance asides. It changes how something is said, never what the rule is — and when it finds a genuine rule conflict it stops and leaves the call to you.
Install: Claude Code
Claude Code reads skills from a .claude/skills/ directory. Project-level scopes them to one codebase; user-level makes them available everywhere:
# Project level (recommended: travels with the repo, so team and agent share one rule set)
git clone https://github.com/Alenkz12/FiboSkills .claude/skills
# Or user level (available in every project)
git clone https://github.com/Alenkz12/FiboSkills ~/.claude/skills
Restart Claude Code (or start a new session) and the fibo-* skills appear in its skill list. Invoke one with /fibo-brainstorming, or just describe your task and let the agent pick the matching stage.
Install: Codex and Other Agents
The skill files are plain Markdown with YAML frontmatter, so they are portable across agents. Clone them wherever you like:
git clone https://github.com/Alenkz12/FiboSkills
- Agents with a native skills mechanism — point their skills directory at the clone, following that agent's own documentation.
- Agents that only read an instructions file — describe the location and purpose of these skills in that file (for example
AGENTS.md) so the agent reads them at the right stage.
Check That They Took Effect
Ask the agent for something small in FiboCode's built-in terminal — "add a config option to this repo", say. When the skills are live it won't edit code first; it will open the brainstorming stage to confirm the AC with you, and then leave spec.md, design.md, and plan.md under .fibo/docs/specs/<feature>/.
Keeping Them Up to Date
Because the skills live in a git clone, pulling the latest is a one-liner:
cd .claude/skills && git pull
Getting Help
- Browse the full skill set at github.com/Alenkz12/FiboSkills.
- Ask questions in GitHub Discussions.
- Report problems at the issue tracker.