mirror of https://github.com/JuliusBrussee/cavekit.git synced 2026-09-14 16:32:40 +08:00

Files

T

Julius Brussee ff0bac1199 v3.0.1: inline-by-default /ck:make; add /ck:make-parallel opt-in

Fix two bugs that could break /ck:make under the quality preset:

1. Caveman mode was telling task-builder subagents to apply caveman-speak
   to their *reasoning*, which corrupted <tool_use> formatting — subagents
   emitted tool calls as body text and terminated with 0 tool_uses.
   Scope caveman to final prose output only (status reports, summaries).
   Reasoning, tool calls, tool arguments, code, commits, and structured
   fields are explicitly excluded. Fixed in make.md, task-builder.md,
   check.md, map.md.

2. Parallel ck:task-builder dispatch with isolation: "worktree" can hit
   a Claude-Code-harness worktree race, returning "[Tool result missing
   due to internal error]" with no agentId. Avoid the race by defaulting
   /ck:make to inline sequential execution: no subagent, no worktree, no
   merge. The parent session implements tasks directly on the current
   branch — the path the user confirmed works.

New command /ck:make-parallel keeps the parallel-subagent path available
as explicit opt-in. It forces TB_ISOLATION=worktree + MAX_PARALLEL=3
(tunable via --concurrency N) for a single run without touching config.
Adds a one-time retry-on-harness-error rule: harness failures re-dispatch
once sequentially, then mark BLOCKED.

Config knob: task_builder_isolation (worktree/inline, default inline).
parallelism defaults lowered: max_agents=1, max_per_repo=1.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-04-18 12:27:20 +02:00

6.1 KiB

Raw Blame History

name, description

name	description
ck-help	Show Cavekit commands and usage

Cavekit

The Loop

/ck:sketch   →   /ck:map   →   /ck:make   →   /ck:check
 what to         how to        build it       verify
  build          build it                       it

Four commands. In that order. That is the whole main cycle.

one-shot:    /ck:ship "<describe your feature>"     # runs all four, no gates

Core Commands

Command	Phase	What it does
`/ck:sketch`	Draft	Decompose your project into domains; write kits with testable acceptance criteria
`/ck:map`	Architect	Read kits, generate a tiered build site with a task dependency graph
`/ck:make`	Build	Inline build loop — parent session implements tasks one at a time (no subagent, no worktree). Safe default.
`/ck:make-parallel`	Build (opt-in)	Parallel subagent build loop — fans out `ck:task-builder` subagents in isolated git worktrees. Throughput at the cost of harness-race surface.
`/ck:check`	Inspect	Gap analysis against kits + peer review of the code; produces APPROVE/REVISE/REJECT verdict
`/ck:ship`	End-to-end	One-shot sketch → map → make → check with no user gates. For tiny features and throwaways.

Auxiliary Commands

Command	What it does	When to use it
`/ck:init`	Bootstrap `context/` and `.cavekit/` state dir; detect capabilities	Once at project start, or re-run after installing a new tool
`/ck:design`	Create / import / update / audit DESIGN.md (visual system)	Before `/ck:sketch` when the project has UI
`/ck:research`	Parallel multi-agent research → research brief in `context/refs/`	Before `/ck:sketch` on novel domains
`/ck:revise`	Trace recent manual fixes back into kits; `--trace` runs the single-failure backprop protocol	After a hotfix or a review finding
`/ck:review`	Branch review — kit compliance + code quality; `--mode gap`, `--codex`, `--tier`, `--strict` narrow the scope	Before merging; as a tier gate inside `/ck:make`; on-demand second opinion
`/ck:status`	Progress report against the build site; `--watch` tails the live dashboard	During or after `/ck:make`
`/ck:config`	Show or update execution preset and runtime keys	Any time
`/ck:resume`	Recover an interrupted loop without losing progress	After a crash, lock conflict, or manual interrupt
`/ck:help`	This page	—

Flags at a Glance

/ck:sketch                        # interactive — asks what to build
/ck:sketch context/refs/          # from reference materials (PRDs, docs)
/ck:sketch --from-code            # reverse-engineer from an existing codebase
/ck:sketch --filter v2            # only v2 kits

/ck:map                           # generate build site from all kits
/ck:map --filter v2               # scope to v2

/ck:make                          # inline sequential build (default, no subagent)
/ck:make --filter v2              # scope
/ck:make --peer-review            # Codex tier gates
/ck:make --max-iterations 30      # iteration cap

/ck:make-parallel                 # opt-in: parallel ck:task-builder subagents in worktrees
/ck:make-parallel --concurrency 4 # set max concurrent packets per wave (default 3)

/ck:check                         # full inspection
/ck:check --filter v2             # scope

/ck:ship "<feature description>"  # one-shot
/ck:ship "add JWT auth" --skip-check
/ck:ship "refactor API" --peer-review

/ck:review                        # two-pass branch review
/ck:review --mode gap             # gap analysis only (replaces /ck:scan)
/ck:review --codex                # Codex-only adversarial review (replaces /ck:judge)
/ck:review --tier                 # tier-gate mode (emits fix tasks back into loop)
/ck:review --strict               # IMPORTANT also blocks
/ck:review --base main            # override diff base

/ck:revise                        # multi-commit revision sweep
/ck:revise --trace                # single-failure backprop (replaces /ck:backprop)
/ck:revise --trace --from-flag    # auto-consume .cavekit/.auto-backprop-pending.json
/ck:revise --trace --from-finding F-012

/ck:status                        # one snapshot
/ck:status --watch                # tail live dashboard (replaces /ck:watch)
/ck:status --watch --interval 5

/ck:init                          # full bootstrap
/ck:init --tools-only             # re-detect CLI tools, MCP servers, plugins (replaces /ck:setup-tools)
/ck:init --tools-only --summary-only

/ck:config                        # show current
/ck:config list                   # show built-in presets
/ck:config preset balanced        # set repo override
/ck:config preset fast --global   # set user default

Model Presets

Preset	Reasoning	Execution	Exploration
`expensive`	opus	opus	opus
`quality`	opus	opus	sonnet
`balanced`	opus	sonnet	haiku
`fast`	sonnet	sonnet	haiku

Precedence: .cavekit/config overrides ~/.cavekit/config, which overrides the built-in default (quality).

Skills (reference docs)

Skill	Topic
`ck:methodology`	Core Hunt lifecycle
`ck:design-system`	How to write and maintain DESIGN.md
`ck:cavekit-writing`	How to write kits with testable criteria
`ck:peer-review`	Cross-model review patterns + Codex loop
`ck:validation-first`	Every requirement must be auto-testable
`ck:convergence-monitoring`	Detecting if loop is converging or stuck
`ck:revision`	Tracing bugs back to kits (includes automated backprop)
`ck:context-architecture`	Organizing `context/` for AI agents
`ck:impl-tracking`	Progress tracking and dead ends
`ck:brownfield-adoption`	Adopting Cavekit on existing codebases
`ck:prompt-pipeline`	Designing prompt sequences
`ck:speculative-pipeline`	Staggered pipeline execution
`ck:documentation-inversion`	Agent-first documentation
`ck:karpathy-guardrails`	Think-before-code, simplicity, surgical changes
`ck:autonomous-loop`	State machine, sentinels, lock protocol

6.1 KiB Raw Blame History