This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

id	title	sidebar_label	last_generated
cheatsheet	Cheatsheet	⚡ Cheatsheet	2026-04-14

Trigger Phrases

Skill	Trigger Phrases
harness-orchestrator	"start a new project", "spin up the harness", "bootstrap a project", "run the harness", "enhance this project", "refactor this codebase", "fix up this app", user provides a full-project prompt
harness-orchestrator (resume)	"resume the harness", "continue the build"
harness-auditor	called by orchestrator (existing projects); or "re-run the auditor"
harness-planner	called by orchestrator; or "re-run the planner", "regenerate a spec"
harness-builder	called by orchestrator; or "re-run QA on the current sprint"
harness-evaluator	called by orchestrator; or "re-run QA on an existing build"
harness-brief	standalone brief generation

Phase Flow

GREENFIELD                              EXISTING PROJECT
─────────────────────────────────────   ──────────────────────────────
Phase 0  Elicit brief                   Phase 0  Elicit brief
           → PROJECT_BRIEF.md                      → PROJECT_BRIEF.md
Phase 1  Init .harness/                 Phase 1A Auditor
           → config.yaml                           → CODEBASE_AUDIT.md
                                                   [GATE: review audit]
Phase 2  Planner                        Phase 2  Planner (delta mode)
           → PRODUCT_SPEC.md                       → PRODUCT_SPEC.md
           [GATE: approve spec]                    [GATE: approve spec]
                ↓ spec approved: touch .graduate, autopilot.enabled: true
Phase 3  Builder (per sprint)           Phase 3  Builder (Sprint 0: housekeeping)
           SPRINT_CONTRACT.md                      then feature sprints
           → code + evaluator                      → code + evaluator
           → QA_REPORT.md                          → QA_REPORT.md
           (fail? retry, max 2)                    (fail? retry, max 2)
Phase 4  Polish (up to 15 iterations)
           Builder refinements → Evaluator → POLISH_REPORT.md
           → append POLISH_HISTORY.md
           (plateau? pivot conditions)
           (all ≥ thresholds → Phase 5)
Phase 5  Final QA Gate → QA_REPORT.md
Phase 6  Completion → HANDOFF.md
           [GATE: user accepts or requests changes]
           (runner mode: write COMPLETE, exit)

config.yaml Keys

Key	Type	Default	Description
harness_version	int	1	Schema version
project_mode	enum	—	greenfield | enhance | refactor | rescue
eval_mode	enum	—	playwright | automated-tests | manual-gates | hybrid
stack	string	pending	Filled by planner or auditor
max_sprint_retries	int	2	Builder/evaluator loops per failed sprint
qa_pass_threshold.functionality	int	7	Sprint pass floor
qa_pass_threshold.design_quality	int	6	Sprint pass floor
qa_pass_threshold.code_quality	int	6	Sprint pass floor
qa_pass_threshold.product_depth	int	6	Sprint pass floor
unattended	bool	false	Set by run-harness.sh; skips all manual gates
manual_gates.after_audit	bool	true	Pause for user after audit (existing projects)
manual_gates.after_spec	bool	true	Pause for user after spec
manual_gates.after_each_sprint	bool	false	Pause for user after each sprint
manual_gates.after_final_qa	bool	true	Pause for user after final QA
polish.enabled	bool	true	Enable iterative polish phase
polish.max_iterations	int	15	Maximum polish iterations
polish.plateau_window	int	3	Iterations to average for plateau detection
polish.plateau_threshold	float	0.5	Avg improvement/iter below this = plateau
polish.pivot_after_stagnant	int	3	Stagnant iterations before pivot eligible
polish.context	enum	pending	frontend | backend | fullstack — set by orchestrator
polish.base_thresholds.*	int	same as qa	Starting thresholds for polish
polish.threshold_step	int	1	Threshold increase per interval
polish.threshold_interval	int	4	Iterations between threshold raises
polish.threshold_overrides	map	{}	Per-criterion step override, e.g. {design_quality: 2}
autopilot.enabled	bool	false	Set when .graduate is written; skips gates
autopilot.design_question_timeout	int	1800	Seconds before auto-default is applied (30 min)
autopilot.interactive_step_limit	int	4	Major steps before checkpointing
autopilot.graduation_trigger	string	spec_approved	Event that writes .graduate

.harness/ Files

File	Writer	Reader	Purpose
config.yaml	Orchestrator + Planner + Auditor	All	Settings, thresholds, mode
PROJECT_BRIEF.md	Orchestrator	Auditor, Planner	User intent, constraints
HANDOFF.md	Orchestrator	Orchestrator (resume)	State: last done, next steps, polish state
PROGRESS.md	Orchestrator (append)	Orchestrator (reconcile)	start/done trail for interrupted-step detection
CODEBASE_AUDIT.md	Auditor	Planner, Builder, Evaluator	Existing codebase analysis
PRODUCT_SPEC.md	Planner	Builder, Evaluator	What to build
CALIBRATION.md	Planner / user	Evaluator	Project-specific score anchors
SPRINT_CONTRACT.md	Builder	Evaluator	This sprint's criteria and builder notes
QA_REPORT.md	Evaluator	Builder (retry), Orchestrator	Sprint scores and feedback
POLISH_REPORT.md	Evaluator (polish)	Builder (polish), Orchestrator	Sub-criteria scores, pivot suggestion
POLISH_HISTORY.md	Orchestrator (append)	Evaluator (drift)	Score history across polish iterations
DESIGN_QUESTION.md	Orchestrator (autopilot)	User, runner	Product/taste decision with options + default
DESIGN_ANSWER.md	User / runner (auto)	Orchestrator	Answer to design question
COMPLETE	Orchestrator	run-harness.sh	Signals successful completion
STOP	User	run-harness.sh	Stop runner after current session
.runner	run-harness.sh	Orchestrator	Runner mode marker
runner.log	Runner + orchestrator	Developer	Session activity log

run-harness.sh Flags

Flag	Default	Description
--max-iterations N	30	Max sessions before stopping
--verbose, -v	off	Stream Claude output to terminal
--watch, -w	—	Colorized live tail of runner.log
--help, -h	—	Show help

Exit codes: 0 = complete, 1 = max iterations hit, 2 = stopped by user, 3 = claude not found

Stop: touch .harness/STOP or Ctrl+C

Resume: re-run the same command — picks up from HANDOFF.md

Cost Reference

Phase	Cost	Duration
Auditor	$1–5	5–15 min
Planner	$0.50–2	3–10 min
Builder (per sprint)	$20–70	30–120 min
Evaluator (per sprint)	$3–5	5–10 min
Polish (per iteration)	$10–25	20–45 min
5-sprint app, 1 retry	$100–200	8–16 h