一键在 Manus 中运行任何 Skill

cavecrew

星标4

分支0

更新时间2026年5月15日 02:23

Use when the user says "delegate to subagent", "use cavecrew", "spawn investigator/builder/reviewer", "save context", or "compressed agent output". Decision guide for delegating to caveman-style subagents (`cavecrew-investigator`, `cavecrew-builder`, `cavecrew-reviewer`) whose tool-results return ~60% smaller than vanilla equivalents, so the main context lasts longer across long sessions.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

danmestas

danmestas/wardrobe

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

Cavecrew

Cavecrew = three subagent presets that emit caveman output. Same job as Anthropic defaults (Explore, edit-style agents, reviewer); difference is the tool-result they return is compressed, so main context shrinks per delegation.

When to use cavecrew vs alternatives

Task	Use
"Where is X defined / what calls Y / list uses of Z"	`cavecrew-investigator`
Same but you also want suggestions/architecture commentary	`Explore` (vanilla)
Surgical edit, ≤2 files, scope obvious	`cavecrew-builder`
New feature / 3+ files / cross-cutting refactor	Main thread or `feature-dev:code-architect`
Review diff, branch, or file for bugs	`cavecrew-reviewer`
Deep code review with rationale + alternatives	`Code Reviewer` (vanilla)
One-line answer you already know	Main thread, no subagent

Rule of thumb: if you'd want the subagent's output in 1/3 the tokens, pick cavecrew. If you'd want prose, pick vanilla.

Why this exists (the real win)

Subagent tool results get injected into main context verbatim. A vanilla Explore that returns 2k tokens of prose costs 2k tokens of main-context budget every time. The same finding from cavecrew-investigator returns ~700 tokens. Across 20 delegations in one session that's the difference between context exhaustion and finishing the task.

Output contracts

What the main thread can rely on per agent:

cavecrew-investigator

<Header>:
- path:line — `symbol` — short note
totals: <counts>.

Or No match. Always file-path-first, line-number-attached, backticked symbols. Safe to grep with path:\d+.

cavecrew-builder

<path:line-range> — <change ≤10 words>.
verified: <re-read OK | mismatch @ path:line>.

Or one of: too-big. / needs-confirm. / ambiguous. / regressed. (terminal first token).

cavecrew-reviewer

path:line: <emoji> <severity>: <problem>. <fix>.
totals: N🔴 N🟡 N🔵 N❓

Or No issues. Findings sorted file → line ascending.

Chaining patterns

Locate → fix → verify (most common):

cavecrew-investigator returns site list.
Main thread picks 1-2 sites, hands paths to cavecrew-builder.
cavecrew-reviewer audits the diff.

Parallel scout (when investigation is broad): Spawn 2-3 cavecrew-investigator calls in one message (different angles: defs vs callers vs tests). Aggregate in main thread.

Single-shot edit (when site is already known): Skip investigator. Hand exact path:line to cavecrew-builder directly.

What NOT to do

Don't use cavecrew-builder when the file isn't already known. Spawn investigator first or the main thread will eat tokens passing context.
Don't chain cavecrew-investigator → cavecrew-builder for a 5-file refactor. Builder will return too-big. and a turn is wasted.
Don't ask cavecrew-reviewer for "general feedback" — it returns findings only, no architecture opinions. Use Code Reviewer for that.
Don't expect prose. Cavecrew output is structured, sometimes terse to the point of cryptic. If a human will read it directly, paraphrase.

Auto-clarity (inherited)

Subagents drop caveman → normal English for security warnings, irreversible-action confirmations, and any output where fragment ambiguity could be misread. Resume caveman after.

Subagent companions

The three subagents this skill orchestrates are bundled separately in the wardrobe as agent components:

agents/cavecrew-investigator — read-only locator
agents/cavecrew-builder — surgical 1-2 file edit
agents/cavecrew-reviewer — diff/branch/file reviewer

Load them alongside this skill to get the full orchestration loop.

同仓库更多 Skills

同仓库

helland

danmestas/wardrobe

The Pat Helland lens: treat durable data as immutable, append-only facts that carry their own identity and version; and design across service, machine, or org boundaries assuming you cannot have distributed transactions — only idempotent, at-least-once messages that must be reconciled. Apply this on in-place updates and overwrites, "just UPDATE the row," two-phase commit or distributed transactions at scale, exactly-once delivery assumptions, shared mutable state across services, and any data that crosses a boundary of service, machine, org, or time. Trigger even when Helland isn't named — the smell is mutation-in-place and pretending distributed systems can be transactional.

2026-07-184

hickey

danmestas/wardrobe

The Rich Hickey lens: pursue simple (un-braided, one role) over easy (familiar, near-at-hand), decomplect concerns that got tangled together, and separate value, identity, state, and time. Apply this whenever "easy," "convenient," or "it's what we already use" is doing the work of justifying a design; when state is braided into logic, policy into mechanism, or data into behavior; when object-soup would be plainer as data; and when a hard design decision is being made at the keyboard under time pressure instead of thought through first. Trigger even when Hickey isn't named — the smell is complecting, and confusing "easy" with "simple."

2026-07-184

carmack

danmestas/wardrobe

The John Carmack lens: minimize the mutable state you have to reason about, do the simple thing before the clever one, and treat correctness tooling — static analysis, warnings-as-errors, determinism — as non-negotiable. Apply this on sprawling mutable state, "we'll clean it up later," debugging by print- statement whack-a-mole, silenced compiler warnings, nondeterministic or non-reproducible behavior, premature abstraction, perf work that skips measurement, and any "what should we actually work on" prioritization call. Trigger even when Carmack isn't named — the smell is code whose full set of states you can't hold in your head, and effort that isn't ruthlessly focused.

2026-07-184

bellard

danmestas/wardrobe

The Fabrice Bellard lens: build the smallest self-contained thing that fully does the job, get performance from insight instead of hardware, and reuse existing machinery instead of reinventing it. Apply this whenever code is growing dependencies, frameworks, services, or infrastructure to solve a problem that a smaller, sharper implementation could handle — or whenever someone says a thing needs a cluster, a big framework, or a rewrite. Use for size/complexity audits, "do we really need this dependency/service?" questions, performance work where the reflex is to add hardware or caching, green-field "should this be one file or a platform?" decisions, and any moment the answer is trending toward "that's impractical, you'd need X." Trigger even when the user doesn't name Bellard — the smell is scope inflation and reaching for resources over insight.

2026-07-184

vault-autoresearch

danmestas/wardrobe

Use when iteratively researching a topic and filing the synthesis into an Obsidian vault for later ingestion. Output goes to `<vault>/raw/` (NOT directly into wiki pages) so that `vault-ingest` can process it on its own schedule. Triggers: '/vault-autoresearch', 'autoresearch [topic]', 'research [topic] into the vault', 'deep dive into [topic]', 'find everything about [topic] for the vault', 'research and stage'.

2026-07-144

vault-ingest

danmestas/wardrobe

Ingest sources into the Obsidian wiki vault. Reads a source, extracts entities and concepts, creates or updates wiki pages, cross-references, and logs the operation. Supports files, URLs, and batch mode. Triggers on: ingest, process this source, add this to the wiki, read and file this, batch ingest, ingest all of these, ingest this url.

2026-07-144

name	cavecrew
version	1.0.0
targets	["claude-code"]
type	skill
description	Use when the user says "delegate to subagent", "use cavecrew", "spawn investigator/builder/reviewer", "save context", or "compressed agent output". Decision guide for delegating to caveman-style subagents (`cavecrew-investigator`, `cavecrew-builder`, `cavecrew-reviewer`) whose tool-results return ~60% smaller than vanilla equivalents, so the main context lasts longer across long sessions.
category	{"primary":"economy"}
license	{"upstream":"MIT","source":"https://github.com/JuliusBrussee/caveman@63a91ec","path":"skills/cavecrew/SKILL.md"}

Cavecrew

When to use cavecrew vs alternatives

Task	Use
"Where is X defined / what calls Y / list uses of Z"	`cavecrew-investigator`
Same but you also want suggestions/architecture commentary	`Explore` (vanilla)
Surgical edit, ≤2 files, scope obvious	`cavecrew-builder`
New feature / 3+ files / cross-cutting refactor	Main thread or `feature-dev:code-architect`
Review diff, branch, or file for bugs	`cavecrew-reviewer`
Deep code review with rationale + alternatives	`Code Reviewer` (vanilla)
One-line answer you already know	Main thread, no subagent

Rule of thumb: if you'd want the subagent's output in 1/3 the tokens, pick cavecrew. If you'd want prose, pick vanilla.

Why this exists (the real win)

Output contracts

What the main thread can rely on per agent:

cavecrew-investigator

<Header>:
- path:line — `symbol` — short note
totals: <counts>.

Or No match. Always file-path-first, line-number-attached, backticked symbols. Safe to grep with path:\d+.

cavecrew-builder

<path:line-range> — <change ≤10 words>.
verified: <re-read OK | mismatch @ path:line>.

Or one of: too-big. / needs-confirm. / ambiguous. / regressed. (terminal first token).

cavecrew-reviewer

path:line: <emoji> <severity>: <problem>. <fix>.
totals: N🔴 N🟡 N🔵 N❓

Or No issues. Findings sorted file → line ascending.

Chaining patterns

Locate → fix → verify (most common):

cavecrew-investigator returns site list.
Main thread picks 1-2 sites, hands paths to cavecrew-builder.
cavecrew-reviewer audits the diff.

Parallel scout (when investigation is broad): Spawn 2-3 cavecrew-investigator calls in one message (different angles: defs vs callers vs tests). Aggregate in main thread.

Single-shot edit (when site is already known): Skip investigator. Hand exact path:line to cavecrew-builder directly.

What NOT to do

Don't use cavecrew-builder when the file isn't already known. Spawn investigator first or the main thread will eat tokens passing context.
Don't chain cavecrew-investigator → cavecrew-builder for a 5-file refactor. Builder will return too-big. and a turn is wasted.
Don't ask cavecrew-reviewer for "general feedback" — it returns findings only, no architecture opinions. Use Code Reviewer for that.
Don't expect prose. Cavecrew output is structured, sometimes terse to the point of cryptic. If a human will read it directly, paraphrase.

Auto-clarity (inherited)

Subagents drop caveman → normal English for security warnings, irreversible-action confirmations, and any output where fragment ambiguity could be misread. Resume caveman after.

Subagent companions

The three subagents this skill orchestrates are bundled separately in the wardrobe as agent components:

agents/cavecrew-investigator — read-only locator
agents/cavecrew-builder — surgical 1-2 file edit
agents/cavecrew-reviewer — diff/branch/file reviewer

Load them alongside this skill to get the full orchestration loop.