← Search

Caveman Learn
Developer tools 🧪 Install ran in our test sandbox · Nov 11, 2026 📌 Pinned version: commit 81958ad
Install test only, on the date shown. This does not guarantee the code is safe.
⚙ Needs: the caveman CLI installed…

Caveman Learn

Act on Caveman token-cost reports — review ranked token sinks and apply cost-lowering fixes with per-edit user consent

At a glance
What
Act on Caveman token-cost reports — review ranked token sinks and apply cost-lowering fixes with per-edit user consent
Cost
Free
Needs
the caveman CLI installed, including `caveman learn` and cavemem: this skill runs `caveman learn report` / `apply` and `caveman mem remember` / `recall`, so without the CLI there is no report to act on. No account or key is needed to install the skill itself; Node.js is used only by the folder's own tests.
Install
Copy the installer prompt below into your Muse — your agent does the rest.

Version:

Quick answer

Caveman Learn is a free Muse skill that acts on Caveman token-cost reports. It reads the JSON report with Cave Score and ranked token sinks, proposes concrete diffs per sink, applies them only with your yes on each edit, verifies the reduction afterward, and reverts if tokens per turn did not drop, never claiming an unmeasured saving.

⌁

Install

Copy the install package below, then paste it into Muse🧪 Install ran in our test sandbox · Nov 11, 2026📌 Pinned version: commit 81958ad

Before you start: tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead. Prerequisites: the caveman CLI installed, including `caveman learn` and cavemem: this skill runs `caveman learn report` / `apply` and `caveman mem remember` / `recall`, so without the CLI there is no report to act on. No account or key is needed to install the skill itself; Node.js is used only by the folder's own tests. Install "Caveman Learn" for me. It is the consent-gated editing half of `caveman learn`: the command measures where an agent's tokens go and ranks the sinks; this skill reviews that report with me and, only with my yes on each individual edit, applies the fixes: trimming a heavy CLAUDE.md, offloading a re-pasted block into caveman memory behind a compact recall pointer, distilling a repeated procedure under a holdout experiment, and memory-health repairs, each fix re-measured afterwards and reverted if it does not actually lower tokens per turn. Source: https://raw.githubusercontent.com/juliusbrussee/caveman/81958ad82f2e2da4417778f81f8a3c838d81b798/skills/caveman-learn/SKILL.md (the skill is the whole skills/caveman-learn/ folder at that commit: SKILL.md plus README.md, CLAUDE.md, package.json and the tests/ folder) Pinned version: install the code at commit 81958ad82f2e2da4417778f81f8a3c838d81b798 exactly (Skill Harbor tested this version). Do not substitute a newer version. If this commit is unreachable, stop and tell me. Updated October 17, 2026: Skill Harbor regenerated this install prompt from the current upstream file. What changed upstream since the version the previous prompt described: how an offloaded block is stored in caveman memory. The skill now writes the exact block, byte for byte, to a temporary file with the file tool and runs `caveman mem remember --stdin < <temp file>`, captures the returned id, and deletes the temp file. The previous file passed the block as a command-line argument (`caveman mem remember -- "<block>"`): the shell would run the block's backticks, expand its variables and end the argument at its first quote. The new file forbids that outright: never put the block on the command line. One correction to the previous prompt itself: it described this skill as a caveman-voice lesson generator. That was wrong; the pinned file is the editing skill described above, and it already was at the previous pin. 1. Download the whole skills/caveman-learn/ folder at the pinned commit (SKILL.md and everything else in the folder) into a temporary folder. 2. Compare SKILL.md with the fingerprint from Skill Harbor's review of 2026-10-17: - SKILL.md: SHA-256 05e803e8024786b4abf78cb89ccbd6aa3834a042f398fc84a69f5652368b2d0a If it does not match, it changed since the review: do not install it yet, tell me what you found and let me decide. 3. Read SKILL.md and tell me in plain words: what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Confirm these points from the file: - Consent is per edit: no "apply all" that hides individual diffs; each fix is proposed as a concrete diff with before -> after tokens/turn, and the net-token-negative gate re-counts after the edit: no drop, revert and report. - The offload storage: the block is located byte for byte and its SHA-256 must match the candidate locator or the item aborts; it is stored via a temp file and `--stdin`, never on the command line, and the temp file is deleted once the memory id is back. - Honest measurement: after = the pointer's tokens/turn plus the recall cost (read from `caveman mem recall "<topic>"`); if after is not below before, the entry is forgotten and the source left untouched; and before finishing, recall must return a hit AND a pointer must be in place, otherwise everything reverts: never make the agent dumber. - Savings are reported by measurement rung (re-counted file, time series, holdout), never summed into one headline; confounders are read out loud; a regression is presented with its verdict and the exact revert path, never softened. - The `caveman learn` analyzer is read-only: this skill is the only writer, and only after a yes; every edit is reversible and reported exactly. Nothing is sent anywhere during installation itself. 4. When I say yes, copy the whole folder's contents into my skills folder, in a folder named "caveman-learn". 5. Check the copy with no network calls (valid frontmatter, files in place) and tell me what was installed, where, and what is left for me to do (for example: have the caveman CLI installed so `caveman learn` can produce the report this skill acts on). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Rules: only touch the temporary folder and the install target. If something seems off, stop and ask me. Never print or paste a secret value.

🔒 Security, provenance and what this touches
🧭 What this touches
Based on the install prompt we scanned, this skill does not run commands, make third-party network calls, install packages, write files, or ask for API keys.
Install prompt analysis · 2026-10-13
🔒 Security and provenance
📌 Pinned version: commit 81958ad · 2026-10-17
🧪 Install ran in our test sandbox · Nov 11, 2026
👁 Observed during the install test · Nov 11, 2026 : registry.npmjs.org
🔎 Last checked · Nov 19, 2026 : no change detected in the files linked by this install prompt
Skill Harbor never reviews the code of a listed skill. These are dated facts about the install prompt and our install test, not a safety certification. Review a skill yourself before you install it. How we verify
Wanna be extra careful?

The install prompt below already includes the vetting steps: your agent follows the community checklist before installing anything with executable code. Want more?

How to check a skill before installing →

✓

Verification certificate

Dated facts about what was checked for this listing, and how. This is not a safety certification.

Pinned version
81958ad · 2026-10-17
Sandbox install test
Install ran in our test sandbox · Nov 11, 2026 · Observed during the install test: registry.npmjs.org
Creator identity
Not verified
Fingerprint watch
Last checked · Nov 19, 2026 : no change detected in the files linked by this install prompt
Install prompt analysis
2026-10-13 : No commands, network calls, packages, file writes or API keys detected.

Skill Harbor never reviews the code of a listed skill. These are dated facts about the install prompt and our install test, not a safety certification. Review a skill yourself before you install it. How we verify

?

Questions

How do I install a skill?

Every product page includes a copy-paste install prompt. Paste it into your Muse and it sets the skill up for you — no manual configuration.

Where does my money go?

Straight to the seller. Skill Harbor never processes payments: checkout happens on the seller’s own page, usually Stripe.

What does the ✓ next to a creator’s name mean?

It means we confirmed the identity of the person behind the listing. It says nothing about the code itself — always check a skill before installing it.