From fakoli-flow
Fast path — skip the full workflow for small tasks under 3 files
npx claudepluginhub fakoli/fakoli-plugins --plugin fakoli-flowThis skill uses the workspace's default tool permissions.
Skip brainstorming, planning, and wave execution for tasks that are too small to justify them.
Designs and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
Designs, implements, and audits WCAG 2.2 AA accessible UIs for Web (ARIA/HTML5), iOS (SwiftUI traits), and Android (Compose semantics). Audits code for compliance gaps.
/flow:quick)Skip brainstorming, planning, and wave execution for tasks that are too small to justify them.
Core principle: One agent, one pass, critic gate, done.
Invocation:
/flow:quick "add timeout param to retry"
/flow:quick "fix import path in auth module"
/flow:quick "rename voiceId to voice_id throughout"
The task description is passed inline. No spec file. No plan file. No waves.
Appropriate for:
Not appropriate for:
If the scope is unclear: estimate it first (Step 1). If the estimate exceeds 2-3 files, stop and suggest /flow:brainstorm instead.
Before dispatching an agent, estimate how many files this task will touch.
Read the relevant files. Look at:
If the estimate is 3 or more files: stop. Tell the user:
This task looks like it will touch 3+ files (<list them>). Quick mode is intended for changes under 3 files.
Suggested path: `/flow:brainstorm` to spec the change, then `/flow:plan` + `/flow:execute`.
To override and use quick mode anyway: `/flow:quick --force "<task>"`
If the estimate is under 3 files: continue.
[ -f tsconfig.json ] && echo "TypeScript"
[ -f Cargo.toml ] && echo "Rust"
{ [ -f pyproject.toml ] || [ -f setup.py ]; } && echo "Python"
This determines the verification command used in Step 4.
Select the agent based on the task type:
| Task type | Agent |
|---|---|
| Code changes, bug fixes, parameter additions | welder |
| Design questions, naming, interface decisions | guido |
| Research, library lookup, API verification | scout |
Default: welder.
If fakoli-crew is installed:
Agent(
subagent_type="fakoli-crew:welder",
prompt="<task description>
Scope: <list the files identified in Step 1>
Language: <detected language>
Make the change. Keep it minimal — only touch what the task requires. Do not refactor unrelated code."
)
If fakoli-crew is not installed: Perform the task directly. Apply the same scope constraint — only touch what the task requires.
After the agent completes (or after making the change directly), run the language-appropriate verification command. Do not skip this step.
TypeScript:
npx tsc --noEmit && bun test
Python:
ruff check . && mypy . && pytest
Rust:
cargo check && cargo test
Read the full output. Check the exit code.
Collect the files the agent modified. Dispatch the critic:
If fakoli-crew is installed:
Agent(
subagent_type="fakoli-crew:critic",
prompt="Review the following files for correctness, style, and completeness:
<list modified files>
Task that was performed: <task description>
Return: PASS, SHOULD FIX (minor issues, non-blocking), or MUST FIX (blocking issues that prevent shipping)."
)
If fakoli-crew is not installed: Review the modified files yourself. Apply the same PASS / SHOULD FIX / MUST FIX judgment.
Report to user:
Done. Task complete.
Files changed: <list>
Verification: PASS (npx tsc --noEmit && bun test — 34/34 passed)
Critic: PASS
Done. No finish step required for quick mode unless the user asks to ship.
One fix cycle only.
Dispatch welder (or fix directly) with the critic's MUST FIX items:
Agent(
subagent_type="fakoli-crew:welder",
prompt="Fix the following issues identified by code review:
<critic's MUST FIX items>
Files: <list>
Do not change anything outside these issues."
)
Re-run verification (Step 4). Re-run critic (Step 5).
If still MUST FIX after one cycle: stop. Report to user:
Quick mode fix cycle did not resolve all issues.
Remaining issues:
<list>
Suggested path: Use `/flow:execute` for a full wave-based fix with multiple review cycles.
Do not loop. One fix cycle is the limit for quick mode.
Log the suggestions. Proceed as PASS. Report both:
Done. Task complete.
Files changed: <list>
Verification: PASS
Critic: PASS (with suggestions logged below)
Suggestions (non-blocking):
- <suggestion 1>
- <suggestion 2>
Quick mode has no spec, no plan, no multi-wave execution, and no sentinel. It is intentionally limited.
If the user asks for something that sounds small but turns out to require a design decision, stop quick mode and say so:
This change requires a design decision: <state the decision>.
Quick mode skips brainstorming. To make this decision properly: `/flow:brainstorm "<topic>"`.
/flow:quick "<task>"
|
v
1. Estimate scope — >3 files? Stop, suggest brainstorm
|
v
2. Detect language (TypeScript / Python / Rust)
|
v
3. Dispatch agent (welder default, guido for design)
|
v
4. Run verification (tsc + bun test | ruff + mypy + pytest | cargo check + cargo test)
|
v
5. Dispatch critic on modified files
|
+-- PASS ---------> Done. Report to user.
|
+-- MUST FIX -----> One fix cycle -> re-verify -> re-critic
| |
| +-- PASS -----> Done.
| +-- MUST FIX -> Stop. Escalate to /flow:execute.
|
+-- SHOULD FIX ---> Log suggestions. Done.