Agent skill

Iii Directory

by iii-hq in iii-hq/workers

Discovery entry point for the engine — search the live function catalog, read the skills, system prompts, and agent profiles that installed workers ship off local disk, browse the public iii workers…

Apache-2.0Auto-check passedAI & LLM Engineering

Install Iii Directory

skills CLI
$ npx skills add iii-hq/workers --skill iii-directory -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install iii-hq/workers iii-directory --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/iii-hq/workers.git skills-src && mkdir -p .claude/skills && cp -r skills-src/iii-directory/skills .claude/skills/iii-directory && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
iii-directory
GitHub stars
113
Token cost
~3.7k tokens
SKILL.md length
1,897 words
Files
5
Skills in repo
19
Repo updated
First seen
Licence
Apache-2.0

At a glance

Discovery entry point for the engine — search the live function catalog, read the skills, system prompts, and agent profiles that installed workers ship off local disk, browse the public iii workers…

  • Works in 2 steps: Register a handler:… → Register the trigger
  • Tasks that involve Prompt engineering
  • SKILL.md covers When to Use, Boundaries, Functions and Reactive triggers
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Iii Directory is an agent skill from iii-hq/workers. Discovery entry point for the engine — search the live function catalog, read the skills, system prompts, and agent profiles that installed workers ship off local disk, browse the public iii workers registry over HTTP, and install new worker bundles. Reach for it first to find out which workers exist and how to call them.

Its SKILL.md is about 3.7k tokens, which your agent loads only when the skill is triggered. The skill folder holds 5 other files (for example `agent-delegation.md`, `function-search.md` and `system-prompts/iii-runtime.md`).

It sits in AI & LLM Engineering, covering Prompt engineering. The licence is Apache-2.0.

When your agent uses it

  • Tasks that involve Prompt engineering

Example prompts

  • “/iii-directory”

Workflow steps

2 steps, taken from the first numbered list in SKILL.md.

  1. Register a handler: registerFunction('my-worker::on-skills-changed', handler).
  2. Register the trigger

What it can do on your machine

Read from SKILL.md and the folder at commit ebfe027. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md (its code samples are typescript).

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Iii Directory loads about 3.7k tokens when it runs. Until then it costs about 84 tokens; SKILL.md has 1,897 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~84
When it runs · the whole SKILL.md, loaded when a task matches
~3.7k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from iii-hq/workers at commit ebfe027, republished under its Apache-2.0 licence (© iii-hq). 1,897 words, ~3,733 tokens.

Download SKILL.mdSave it as .claude/skills/iii-directory/SKILL.md (or your agent's skills folder). This skill also uses 4 other files; get the full folder from GitHub.
name
iii-directory
description
Discovery entry point for the engine — search the live function catalog, read the skills, system prompts, and agent profiles that installed workers ship off local disk, browse the public iii workers registry over HTTP, and install new worker bundles. Reach for it first to find out which workers exist and how to call them.

iii-directory

The directory worker is how an agent finds its way around the engine. It exposes five surfaces: function search (directory::search_functions), installed-worker docs (directory::skills::*), chat identity prompts (directory::system-prompts::*), reusable agent profiles (directory::agents::*), and the public worker catalogue (directory::registry::*). A download pulls a bundle onto disk, and each filesystem-backed family also takes direct create / update / delete calls — those are this worker's only writes. Everything else here is read-only.

Two kinds of id flow through this worker and they must not be mixed up. A callable id uses :: (directory::skills::get) and goes in the function: field of agent_trigger. A skill id uses / (iii-sandbox, agent-memory/observe) and names a document — pass it as the id argument to directory::skills::get. The ids that list and index print are skill ids; a worker's overview is the bare worker name (iii-sandbox, not iii-sandbox/index, and the iii- prefix is never dropped). Use the id you were given — do not invent one.

Only installed workers are visible. index, list, and get show on-disk skills for installed workers, plus this worker and the iii engine which are always present. A skill you know exists stays invisible until its worker is downloaded, so when one is missing, install it and look again. With auto_download enabled the worker subscribes to the engine worker add event and pulls a newly added worker's skills automatically, so freshly installed workers can appear without a manual download. System-installed agent skills under the read-only agents_skills_folder (.agents/skills under III_COMPOSE_DIR, or the process current directory when standalone) are also always visible in list/get — they are skills, not workers, so they never appear in index and update/delete refuse them.

When to Use

  • You need functions for a step's unmet capabilities (usually one to six, at most eighteen) — directory::search_functions.
  • You need to see which workers are installed — directory::skills::index (token-light; start here).
  • You need to read a worker's overview or a deeper doc it linked to — directory::skills::get.
  • You need to find a skill across the repo with filters — directory::skills::list.
  • You need the system prompts the chat picker offers as an identity override — directory::system-prompts::list / get.
  • You need a reusable agent profile (display name, emoji logo, skill selection, and its system prompt) — directory::agents::list / get.
  • You are deciding whether to install a worker — directory::registry::workers::info is the pre-install card: public function/trigger names with descriptions, config, dependencies, skill paths (no schemas; readme: true adds the README). Contracts come from engine::functions::info after install.
  • You need to install a published worker's skills — directory::skills::download_from_registry.
  • You can only reach the directory:: namespace but need one engine function's exact schema — directory::engine::functions::info.

Boundaries

  • Only installed workers are visible. If the engine daemon is unreachable at boot, filtering is skipped and everything on disk is shown instead.
  • Writes are downloads plus the per-family create / update / delete calls; every read function leaves disk untouched.
  • Skills under agents_skills_folder are read-only: update/delete refuse them (D116), and create refuses ids in their namespaces (D115). Edit them with their owning tool, or copy one into skills_folder on disk to fork it.
  • Not the live-connection view. directory::* reflects what is on disk or in the registry, not what is connected right now. For that, call the engine directly (engine::functions::list, engine::workers::list, …); daemon-managed providers (http, cron, state) open no WebSocket, so merge worker::list by name.
  • Do not put a skill id (/) in agent_trigger's function: field, and do not pass a function id (::) to directory::skills::get.
  • System prompt files without a description: in frontmatter are silently skipped by directory::system-prompts::list.
  • Skills and system prompts share skills_folder; a system-prompts/ path component selects the prompt family, other prompts/ paths are ignored, and agents/ is reserved. Agent profiles are direct <agents_folder>/<id>.md files.
  • An agent profile's skills: list is its PRELOADED skills, the twin of functions:: the harness freezes each skill's body into the prompt of every session running as the profile (parents' lists included, root-first). It grants no access and never narrows the skills index; unknown ids are reported by get as unknown_skills warnings rather than errors and rendered as unavailable. agents_folder profiles are unrelated to the read-only agents_skills_folder (~/.agents/skills), which holds external tools' skills.
  • Registry answers (registry::workers::list / info) are cached ~60 s per unique input by default (registry_cache_ttl_ms) — change a parameter to refresh.

Functions

  • directory::search_functions — find compact installed and installable function candidates for the required capabilities (usually one to six, at most eighteen per call).
  • directory::skills::index — token-light per-worker overview, one block per installed worker; truncates and tells you to call list when large.
  • directory::skills::list — enumerate every visible skill with id/title/type/description/bytes/modified_at; narrow with search, prefix, type, or include_description.
  • directory::skills::get — read one skill doc by its skill id; forgiving about short names, a trailing .md, an iii:// prefix, and SKILL.md filenames. The response's path is the absolute on-disk file; its parent directory is the skill's base directory, where payload the body references by relative path (scripts/, reference/) lives.
  • directory::skills::download_from_registry — install a published worker's skills from the registry; worker required, pin with version XOR tag (default tag: latest).
  • directory::skills::download_from_repo — pull one skill folder from a GitHub repo; repo + skill required, branch defaults to main.
  • directory::skills::download — flexible alias accepting either source set; prefer the two explicit forms so the source is unambiguous.
  • directory::skills::update — overwrite one EXISTING skill with new full-file markdown (frontmatter included); never creates — author with directory::skills::create or materialize a bundle with a download first.
  • directory::skills::create — create a NEW skill at <skills_folder>/<id>.md from full-file content; refuses an id that already resolves in the visible set, an existing target path, and ids the visibility filter (or a system-installed agents namespace) would hide.
  • directory::skills::delete — permanently remove one EXISTING skill by id (same forgiving id forms as get); cleans up parent directories left empty.
  • directory::system-prompts::list — list the system prompts the chat picker offers; the response array is named prompts.
  • directory::system-prompts::get — read one system prompt's body by name; raw: true also returns the full on-disk file for round-tripping.
  • directory::system-prompts::create — create a NEW system prompt at <skills_folder>/system-prompts/<name>.md; refuses an existing name or target path.
  • directory::system-prompts::update — overwrite one EXISTING system prompt; the frontmatter must keep a non-empty description (a declared name renames it).
  • directory::system-prompts::delete — permanently remove one EXISTING system prompt by name.
  • directory::agents::list — list agent profiles (id, display name, description, emoji logo, extends, skill_count where null means every skill, function_count of preloaded functions; model/reasoning_effort are resolved through extends; skill_count/function_count count the additive union of the chain's lists). The bundled profiles default (the compact directory-first identity, shown as "Default"), iii (the harness default identity, hidden from the gallery), and iii-minimal (a hidden alias that extends default) are always listed with builtin: true; composer_placeholder, when present, is the profile's own example for the empty chat composer and never inherits; a row with inheritance_error has a broken extends chain.
  • directory::agents::get — read one agent profile by id: its RESOLVED system prompt (ancestors' bodies root-first via extends, then its own; blank bodies are skipped, so a prompt-less profile serves its parent chain, or "" when it has no parent), preloaded skills (skill ids whose bodies the harness pre-loads into the session prompt) + unknown_skills, preloaded functions (engine function ids whose contracts the harness pre-loads into the system prompt of every session running as this profile) + unknown_functions (ids the engine does not know right now — a warning), and model (the profile's default model id — use it when spawning/sending with this profile; null = caller decides); raw: true returns the profile's OWN file for editing. inheritance_error set = fix extends before running it.
  • directory::agents::create — create a NEW agent profile at <agents_folder>/<id>.md from full-file content; frontmatter needs a non-empty name, logo is emoji-only, skills: and functions: (preloaded function ids, e.g. coder::tree) are optional lists, and the body — the system prompt — may be empty; an extends: <id> that does not resolve is not a write error — get reports it as inheritance_error.
  • directory::agents::update — overwrite one EXISTING agent profile (same scanner rules; the id stays the file stem, frontmatter name is display-only). Updating the bundled iii creates the local file that shadows it.
  • directory::agents::delete — permanently remove one EXISTING agent profile by id; running sessions are unaffected, profiles extending it stop resolving until fixed. Deleting a local iii falls back to the bundled copy.
  • directory::agents::functions::add / directory::agents::functions::remove — { id, functions: ["coder::tree", …] }: add ids to / drop ids from one EXISTING profile's OWN functions: list without a raw round-trip; the rest of the file stays byte-identical, ids already present (or already absent) are ignored, unchanged: true means nothing was written. Editing a bundled profile creates the local shadow. Ids inherited through extends are not touched: the resolved list is the union of the chain (root first), so an inherited id is removed on the parent that declares it.
  • directory::registry::workers::list — page through published workers in the public registry (pagination.next_cursor feeds the next page's cursor).
  • directory::registry::workers::info — pre-install card for one worker, including ones not installed: envelope, api_reference (public functions + triggers, names and descriptions only) and skills_tree; readme: true adds the README.
  • directory::engine::functions::info — thin proxy to the engine's engine::functions::info; returns request/response schema, metadata, and registered triggers for one function id.
Show full SKILL.md (438 more words)Show less

A failed call returns one plain sentence carrying a Did you mean: suggestion and a Next: function to call (codes D110/D112/D210/D310/D311, D410 for a missing agent profile, D416 for an agent functions::add/remove request with no valid function ids, D320 when the registry is unreachable, and on the write paths D213 for content the next scan would skip, D214/D114/D414 for a create whose name/id or target path is already taken, D115 for a skill id the visibility filter or an agents namespace reserves, and D116 for a write to a read-only system-installed skill) — follow it instead of retrying the same input. Downloads overwrite file-by-file, so hand-edited extra files survive a re-pull.

Reactive triggers

The worker publishes three custom trigger types, one per kind — directory::skills::on-change, directory::system-prompts::on-change, and directory::agents::on-change. Each fires for its own kind only, on any of: a download that wrote at least one file of that kind (op: "download"), that family's update, create, or delete (op: "update" / "create" / "delete"), or a change made to that kind's files directly on disk, outside this worker (op: "external" — a file pasted in, edited in an external editor, deleted, or renamed). Bind one when a different worker must react to the on-disk set changing; the mcp worker uses this to emit notifications/*_list_changed to its clients without re-polling. Direct <id>.md profile edits under agents_folder fire directory::agents::on-change; nested files there are ignored.

Reach for it when:

  • A worker caches the skill, system-prompt, or agent-profile list and must invalidate it on change.
  • You want a push the moment new bundles install, instead of polling directory::skills::list.

Do not bind when:

  • You ran the download yourself — its return payload already lists skills_written / system_prompts_written / agents_written.
  • Your own reaction writes .md files under skills_folder through some path OTHER than this worker's update / create (a shell or coder worker, say). Writes made through this worker are suppressed, but outside writes are not: your handler would re-trigger itself.
How to bind
  1. Register a handler: registerFunction('my-worker::on-skills-changed', handler).
  2. Register the trigger:
typescript
iii.registerTrigger({
  type: 'directory::skills::on-change',
  function_id: 'my-worker::on-skills-changed',
})

Delivery is fire-and-forget (best-effort, at-most-once): a slow or failing subscriber is logged and skipped so it cannot block the write path. Direct edits under skills_folder DO fire it now, as op: "external" — a filesystem watch supplies that, coalescing a burst into one event per kind. Read calls never fire it, and neither does this worker's own writing twice (a create or update sends its precise op, not an extra external). The watch is a doorbell, not a ledger: every read re-scans disk, so a missed event costs a stale open view until the next call, never data. For the event payload shape, call get function info on the trigger type.

© iii-hq, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 4 other files in iii-directory/skills of iii-hq/workers.

  • SKILL.md
  • agent-delegation.md
  • function-search.md
  • system-prompts/iii-runtime.md
  • worker-microvm-service.md

Open the folder on GitHubat commit ebfe027

Compare with similar skills

Iii Directory next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Iii Directory compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Iii Directory this skilliii-hq/workers113—~3.7kAutomated safety check: PassApache-2.0
Prompt Improverseverity1/claude-code-prompt-improver1.9k1 repos~1.7kAutomated safety check: PassMIT
Prompt Engineering Patternsynulihao/AgentSkillOS61814 repos~1.7kAutomated safety check: PassNone
Patch CreationPiebald-AI/tweakcc2.5k—~1.6kAutomated safety check: PassMIT
Senior Prompt Engineermaslennikov-ig/claude-code-orchestrator-kit2603 repos~1.4kAutomated safety check: PassCustom licence
Codex Fable5baskduf/FableCodex437—~1.6kAutomated safety check: PassAGPL-3.0

Similar skills

  • Prompt Improver

    severity1/claude-code-prompt-improver

    This skill enriches vague prompts with targeted research and clarification before execution.

    1.9k GitHub starsUsed in 1 repo~1.7k tokens
    AI & LLM EngineeringAuto-check passed
  • Prompt Engineering Patterns

    ynulihao/AgentSkillOS

    Master advanced prompt engineering techniques to maximize LLM performance, reliability, and controllability in production.

    618 GitHub starsUsed in 14 repos~1.7k tokens
    AI & LLM EngineeringAuto-check passed
  • Patch Creation

    Piebald-AI/tweakcc

    Create and register new patches for tweakcc. An agent skill from Piebald-AI/tweakcc.

    2.5k GitHub stars~1.6k tokensUpdated today
    AI & LLM EngineeringAuto-check passed
  • Senior Prompt Engineer

    maslennikov-ig/claude-code-orchestrator-kit

    Provides reference guides and Python scripts for prompt optimization, RAG evaluation, and agent orchestration when building or tuning LLM systems.

    260 GitHub starsUsed in 3 repos~1.4k tokens
    AI & LLM EngineeringAuto-check passed
  • Codex Fable5

    baskduf/FableCodex

    Apply a Claude Fable 5 inspired operating style inside Codex.

    437 GitHub stars~1.6k tokensUpdated 2 mo ago
    AI & LLM EngineeringAuto-check passed
  • Reference for designing and tuning production LLM prompts: few-shot examples, chain-of-thought, structured outputs, templates and system prompts.

    40k GitHub stars~1.3k tokensUpdated 5 days ago
    AI & LLM EngineeringAuto-check passed

More from iii-hq/workers

All 19 skills in this repo
  • Cron

    iii-hq/workers

    Schedule any registered function on a 6- or 7-field cron expression with the standalone cron worker.

    113 GitHub stars~1.2k tokensUpdated today
    Auto-check passed
  • Email

    iii-hq/workers

    Send and read email from the iii engine — SMTP send, IMAP read, and real-time IDLE push as a subscribable trigger type.

    113 GitHub stars~2.1k tokensUpdated today
    Auto-check passed
  • HTTP

    iii-hq/workers

    Expose registered functions as HTTP endpoints with the standalone http worker.

    113 GitHub stars~1.3k tokensUpdated today
    Auto-check passed
  • Kanban

    iii-hq/workers

    File-backed kanban board: create, read, move and comment on tickets by key or uuid, assign agent profiles, and wake on comments through the kanban:comment trigger instead of polling.

    113 GitHub stars~1.4k tokensUpdated today
    Auto-check passed
  • Pubsub

    iii-hq/workers

    DEPRECATED (will be removed in an upcoming release; migration guide https://iii.dev/docs/upgrading/migrate-from-streams).

    113 GitHub stars~1.2k tokensUpdated today
    Auto-check passed
  • Release Sync

    iii-hq/workers

    Organize worker release tags into Linear release waves — one team document plus one release/<date label per same-day batch of worker releases, with shipped MOT issues labeled and linked.

    113 GitHub stars~1.1k tokensUpdated today
    Auto-check passed

Questions about Iii Directory

What does Iii Directory do?

Discovery entry point for the engine — search the live function catalog, read the skills, system prompts, and agent profiles that installed workers ship off local disk, browse the public iii workers…. Iii Directory is an agent skill from iii-hq/workers. Discovery entry point for the engine — search the live function catalog, read the skills, system prompts, and agent profiles that installed workers ship off local disk, browse the public iii workers registry over HTTP, and install new worker bundles.

When should I use Iii Directory?

Iii Directory fits situations like: tasks that involve Prompt engineering.

How do I install Iii Directory in Claude Code?

Run `npx skills add iii-hq/workers --skill iii-directory -a claude-code`. Or copy the skill folder (iii-directory/skills in iii-hq/workers) into .claude/skills/iii-directory in your project. Claude Code loads it when a task matches its description.

How do I install Iii Directory in Codex?

Run `npx skills add iii-hq/workers --skill iii-directory -a codex`. Or copy the skill folder (iii-directory/skills in iii-hq/workers) into .agents/skills/iii-directory in your project. Codex loads it when a task matches its description.

Can I use Iii Directory in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add iii-hq/workers --skill iii-directory -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/iii-directory, .gemini/skills/iii-directory, .github/skills/iii-directory and .opencode/skills/iii-directory in your project.

What does Iii Directory need to run?

SKILL.md names no scripts, command-line tools or credentials: Iii Directory is instructions for the agent only.

Does Iii Directory access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Iii Directory safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Iii Directory use?

Iii Directory is published under the Apache-2.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Iii Directory use?

About 3.7k tokens (SKILL.md is roughly 15k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Iii Directory?

Skills that share tags, products or a category with Iii Directory: Prompt Improver (severity1/claude-code-prompt-improver, 1.9k stars), Prompt Engineering Patterns (ynulihao/AgentSkillOS, 618 stars), Patch Creation (Piebald-AI/tweakcc, 2.5k stars) and Senior Prompt Engineer (maslennikov-ig/claude-code-orchestrator-kit, 260 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Iii Directory?

iii-hq (a GitHub organization) maintains it in iii-hq/workers, which has 113 GitHub stars. The repository holds 19 skills in this directory. The repository was last updated on October 10, 2026.

Source: iii-hq/workers on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.