Skip to content
Agentmart
Submit a tool

caveman

Caveman form for internal agent traffic (briefs, agent returns, research fan-outs, workflow returns, tool-call notes).

Skill by plurano

IndexedEUPL-1.2
Write a review

Chalk Labs

Need help deploying caveman?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/plurano-caveman

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Caveman form for internal agent traffic (briefs, agent returns, research fan-outs, workflow returns, tool-call notes). Fragments, one fact per line, code, paths, errors verbatim, nothing a reader pays for twice. The `internal` text register; forge text uses social-text, docs and human replies use full prose.

Use cases: research, productivity, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/caveman

What's inside

Skill name
caveman
Description
Caveman form for internal agent traffic (briefs, agent returns, research fan-outs, workflow returns, tool-call notes). Fragments, one fact per line, code, paths, errors verbatim, nothing a reader pays for twice. The `internal` text register; forge text uses social-text, docs and human replies use full prose.
License
EUPL-1.2
Metadata
derived_from: https://github.com/JuliusBrussee/caveman (MIT)
Instructions
137 lines in SKILL.md · view on GitHub

Files (2)

  • LICENSE
  • SKILL.md

SKILL.md (first lines)

# Caveman: internal register (`caveman`)

Adapted from [Caveman](https://github.com/JuliusBrussee/caveman) by Julius
Brussee and
[contributors](https://github.com/JuliusBrussee/caveman/graphs/contributors);
MIT terms at adaptation (2026-09-18).

Caveman = `internal` register of text-register policy (`register:` in
`.standards.yaml`; "Text Register" section of AGENTS.md). Config value stays
`internal`; this skill = its form. Agent traffic pays per token, per hop ->
caveman line carries same facts, fewer tokens.

## Scope

Caveman covers text another agent reads:

- briefs, task prompts to subagent or cheap lane;
- agent returns, workflow returns;
- research fan-out results;
- tool-call notes, progress lines between agents.

Human-facing text keeps its own register: forge text (issues, PR bodies, review
comments, commit bodies, changelog titles) -> `social-text`; documentation tree
(`docs`), READMEs, ADR bodies -> docs register; reply to human operator -> full
prose.

## Rules

1. **Cut grammar with no fact.**

   | Drop | Examples |
   | :--- | :--- |
   | articles | `a`, `an`, `the` |
   | pronouns | `I`, `we`, `it` |
   | copulas | `is`, `are`, `was` |
   | auxiliaries | `have`, `will`, `did` |
   | hedges | `seems`, `probably`, `might` |
   | politeness | `please`, `thanks` |
   | framing | `I think`, `note that`, `it looks like`, `as requested` |

2. **Fragments.** One fact per line. Lists over sentences.
3. **Symbols** replace connective phrases:

   | Symbol | Reads as | Example |
   | :--- | :--- | :--- |
   | `->` | causes, leads to, then | `stale receipt -> push rejected` |
   | `=` | is, equals, means | `block = 15 lines` |
   | `x2` | count, times | `retry x3` |
   | `!` | warning, risk | `! gate not rerun after rebase` |
   | `?` | open question | `? Windows path untested` |

4. **Verbatim tokens.** Code, paths, `file:line`, commands, flags, error text,
   ids, shas, version strings, numbers = copied exact, never abbreviated.
   Compression hits English around them, never them. Invented abbreviations
   (`cfg ldr`, `ctx cmp`) cost reader more decoding than they save; use full
   word or real identifier.
5. **Start with answer.** No restated brief, no step recap before result, no
   closing line. Verdict first, then evidence.
6. **Tables only** when more than 3 rows compare more than 2 fields. Otherwise:
   list.
7. **Return shape** follows register block: verdict, changed paths, commands
   run, evidence pointers, open questions.
8. **Evidence** follows register block, same wording: evidence above 58 lines or
   1500 tokens -> file in the directory the register block names, which the
   repository's register policy can configure; return
   `evidence: <path> sha256:<12 hex> lines:<n>`, fetch only when decision needs
   it. Manifest may tighten both numbers; block in AGENTS.md prints values in
   force.

## Static checker

`praetorctl caveman check` defaults to runtime message form. Exception:
AGENTS.md + compiled vendor files at own path below `--root` (exact,
case-sensitive) default to `--kind=context`, same verdict as gate. Explicit
`--kind` wins. Structured traffic selects kind explicitly:

- brief: `--kind=brief`; fields = `goal`, `inputs`, `return`, `evidence`,
  `task`; goal first;
- return: `--kind=return`; fields = `verdict`, `changed`, `ran`, `evidence`,
  `open`; verdict first;
- policy text: `--kind=context`; C1-C8 heuristics only, uncovered skill rules
  reported advisory.

PASS summary lists mechanically checked and advisory skill-rule numbers. Runtime
producer wiring separate; checker validates supplied text only.

## Clarity floor

Caveman = shorter text, same information, never shorter text that makes reader
guess. Before sending: check every line -> can reader act on it without asking
back? When compression drops fact reader needs (which file, which count, new
failure or pre-existing, check bypassed or not), put fact back. Keep negations
explicit (`no bypass`, `not rerun`); dropped `not` inverts line.

## Before and after

Real return from a praetor lane:

<!-- caveman:off -->
- Before: "First two pushes were rejected by the pre-push gate with 'state
  synchronization stale' because gate run wrote .standards-receipt.json after
  the last state sync; fixed by re-running state sync and pushing again. Not
  bypassed."
<!-- caveman:on -->
- After: `push rejected x2: state stale (gate run wrote receipt after sync).
  fix: resync, push. no bypass.`

Test return:

<!-- caveman:off -->
- Before: "I ran the tests for the repository with the race detector and all of
  them passed. go vet did not report any issues. The only file I changed was
  .agents/skills/caveman/SKILL.md, where I added an error for a missing end
  marker."
<!-- caveman:on -->
- After:

  ```text
  verdict: pass

From plurano/peritia, EUPL-1.2.

Signals

Stars
0
Forks
0
Open issues
1
Last commit (this folder)
Oct 18, 2026 (12 days ago)
Last release
Oct 15, 2026 (15 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
EUPL-1.2
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

Same skill elsewhere

Copies of this SKILL.md published by others; search shows only one of them.

More from plurano

Done for you

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped
50+
launches shipped
media placements
500+
media placements
founders on every account
2
founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get caveman running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Goes to the Chalk Labs founders, not the newsletter. Privacy

Need help deploying caveman?

CallGet help
Cite this page

caveman skill. Agent Mart. https://agentmart.dev/skills/plurano-caveman (last updated 2026-10-24)