A library of curated agent skills, bundled with Aletheore

Skills your agent can read right now.

Searching the catalogue needs no download, no sign-in and no network call. Hallbook lives inside Aletheore's MCP server. Your agent describes the task once and gets the right skill back, or an honest “nothing here fits”. This box replays recorded answers from the real ranking.

Request sliprecorded examples

try:

NONELOWHIGH

Nothing to install per skill, per agent

Skills usually travel as folders: find one, copy it into each agent's skills directory, restart the agent, and do it again for the next skill, the next agent and the next machine. Hallbook replaces all of that with one connection.

Connect Aletheore once, as you would any MCP server, then ask your agent to use the skills. Nothing to copy, no plugin to install, no restart, no setup per agent.

One call, then your agent moves on

Agents lose time searching for the right skill, then searching again. Hallbook answers in a single call, and when it is not sure it says so, so the agent carries on without a wrong guide in its head. This is a recorded call against the real catalogue.

~/.claude/skills~/.codex/skills~/.agents/skillsfolders we never write to

Your own skills live here too

Write a skill once, or ask your agent to save what you just worked out, and it sits next to the catalogue. Every agent connected to Aletheore finds it the same way, so there are no folders to keep in step between tools.

1
Add one from a file
```
aletheore hallbook add my-skill ./my-skill.md
```
Name it plainly. Custom skills are found by name, and the original is kept as a .bak if you add it again.
2
Let your agent save it
```
Save what we just figured out as a skill called deploy-checklist.
```
The agent calls aletheore_add_custom_skill. Next session, any agent can find it.
3
Take it with you
```
aletheore login
```
Sign in once and your custom skills follow your account to your other machines.

Custom skills show up in search and in the one-call lookup next to the 1,129 catalogue ones, with ids like custom/deploy-checklist. They are stored on your machine. Once you sign in, each one you add or change is also sent to your Aletheore account, stored there as plain text, and synced to your other machines, so keep secrets out of them.

Six shelves

Chosen for work that agents actually do. Counts are read from the catalogue.

Skills per shelf

Build528platforms · backend · frontend · mobile
AI & Agents139ai-engineering · agent-workflows · browser-automation
Ship & Operate137devops · documentation · git-workflow · reliability
Content & Media115marketing · video-media · documents
Quality114security · testing · refactoring · code-review
Product & Design96product · design

What we check before a skill ships

A skill is instructions your agent will follow, so a bad one is a real risk. This is a real skill file from the catalogue. The list is what our checks cover, and where they stop.

---name: systematic-debuggingdescription: Use when encountering any bug, test failure, or unexpected behavior, before pro…skill_id: debugging/systematic-debuggingtitle: Systematic Debugging1source_url: https://github.com/obra/superpowers/tree/bb92a77741419a4ab5f06e711a283343f1ada0c…2license: MITshelf: quality4aletheore: {"id": "debugging/systematic-debugging", "shelf": "quality", "category": "debuggi…--- 3# Systematic Debugging ## Overview **Core principle:** ALWAYS find root cause before attempting fixes. Symptom fixes are failur… **Violating the letter of this process is violating the spirit of debugging.** ## The Iron Law ```NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST``` If you haven't completed Phase 1, you cannot propose fixes. ## When to Use Use for ANY technical issue:- Test failures- Bugs in production- Unexpected behavior- Performance problems

1Pinned source
The exact repository and commit it came from, and the author is credited. Nothing is fetched from the network at read time.
2Its own licence
Each skill carries its licence. Only permissive ones (MIT, Apache-2.0, BSD, ISC and similar) are accepted.
3Instruction pattern lint
Instruction text is matched against two known-bad shapes: a download piped straight into a shell (curl or wget into sh, bash and similar), and phrasing that asks to send a secret or credential to a URL. It is a lint, not a general prompt-injection detector, and a clean result is not a security review.
4Curation record and bundled scripts
Shelf, quality score and the vetting result sit next to the skill. A bundled scripts/ folder is run through our secrets scanner, and only that: the pattern lint above reads the instruction text, not the scripts. A skill that fails any check is held back, and a person reviews each finding before it is cleared.

53: hits from a broader curation-time scan for override phrases, piped shell commands and hidden characters. All turned out to be documentation or benign examples.
6: candidates held back by the scanner, then reviewed one by one. Most were false alarms. One was left out because its bundled script sends usage events to a third-party analytics service.
0: skills Aletheore writes into your agent's folders. You read them through the MCP tool or the CLI, and the AGENTS.md line and hook config are printed for you to paste.

Receipts

We measured it, including the parts that did not go our way. These are our own tests on this catalogue, October 2026.

Finding the right skill from a vague request

994 requests written by a different model, in the voice of someone who does not know the skill names.

Finding the right skill from a vague request

First result is the right skill45%before9%
Right skill in the top five73%before21%

When it says “high”, it is right

The lookup only hands over a skill when confident. Otherwise it gives short candidates, or says nothing fits.

Precision of a confident answer

Confident answer is right85 to 89%

Getting your agent to use it

Share of runs that called Hallbook. “Say it” means one line in your prompt: Use the Aletheore skills available to you.

Share of runs that used Hallbook

You say it in the prompt96 to 100%
One line in AGENTS.md12 to 54%
Nothing said0 to 4%

The same question on other agents, counted in runs that called Hallbook. Small samples, so read the direction, not the decimals. “Snippet” is the one line in the agent's own instructions file.

AGENT	NOTHING SAID	SNIPPET	SAY IT
Claude Code (subagents)	0 of 6	5 of 6	5 of 6
Antigravity CLI	0 of 6	5 of 6	6 of 6
Codex CLI	0 of 12	3 of 12	2 of 12

Use it

Install it. Then just ask your agent to use the skills.

1 · INSTALL AND START THE SERVER

pip install aletheore

aletheore mcp .

2 · SAY IT

Use the Aletheore skills available to you for this task.

3 · OR MAKE IT AUTOMATIC (CLAUDE CODE PROMPT HOOK)

aletheore hallbook hook

Built on MCP, so any MCP agent can call it. We have tried it with OpenCode, Antigravity CLI, Codex CLI and Claude Code subagents. Agents rarely reach for it on their own, so add the line from step 2 or 3 to your agent's instructions file; most runs on Antigravity and Claude used it that way, and fewer on Codex. Your own skills can be added once and used everywhere. Every skill credits its author and links to its source.

Nothing to install per skill, per agent

One call, then your agent moves on

Your own skills live here too

Add one from a file

Let your agent save it

Take it with you

Six shelves

What we check before a skill ships

Receipts

Finding the right skill from a vague request

When it says “high”, it is right

Getting your agent to use it

Install it. Then just ask your agent to use the skills.