Skip to content
Skills Library
AI & agents · Building & coding

create-skill

/create-skill

[Orbit 3.3.3 Docs & writing] Create or improve a skill and wire it into the Orbit pipeline. Use when: «сделай скилл», «добавь в пак», «новый навык для агента», «улучши скилл»; create skill, improve an existing skill.

Licence
MIT
Last repository change
09.11.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
18
Scripts
10
HTML comments
0
URLs in SKILL.md
1
Deprecation

No signals detected. Repository changed on 09.11.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What to check before installing
  • 10 scripts in this package. Review them before installing, as Claude may execute them.

Scripts: tools/eval-viewer/generate_review.py, tools/scripts/__init__.py, tools/scripts/aggregate_benchmark.py, tools/scripts/generate_report.py, tools/scripts/improve_description.py, tools/scripts/package_skill.py, tools/scripts/quick_validate.py, tools/scripts/run_eval.py, tools/scripts/run_loop.py, tools/scripts/utils.py

What it does and when to use it

[Orbit 3.3.3 Docs & writing] Create or improve a skill and wire it into the Orbit pipeline. Use when: «сделай скилл», «добавь в пак», «новый навык для агента», «улучши скилл»; create skill, improve an existing skill.

Original author description

[Orbit 3.3.3 Docs & writing] Create or improve a skill and wire it into the Orbit pipeline. Use when: «сделай скилл», «добавь в пак», «новый навык для агента», «улучши скилл»; create skill, improve an existing skill.

How it works

  1. When to use
  2. Inputs
  3. Procedure
  4. Output
  5. Check
  6. Example
  7. Pitfalls
  8. Portability
  9. What's improved vs the source
  10. Based on
Original SKILL.md headings · 128 lines, 1,241 words
SKILL.md preview
---
name: create-skill
description: '[Orbit 3.3.3 Docs & writing] Create or improve a skill and wire it into the Orbit pipeline. Use when: «сделай скилл», «добавь в пак», «новый навык для агента», «улучши скилл»; create skill, improve an existing skill.'
---

# create-skill — create or improve a skill

> **Orbit · ③ Operate · step 3.3.3 Docs & writing**  
> **Input:** a recurring task ← `instruct-agents`  
> **Output:** a new skill → next `write-update`  
> **Works with:** `instruct-agents`  
> **Shared rules:** reply in the user's language per `speak-plain`; generated content defaults to English; stay in scope; artifacts go to `.orbit/` (contract: `ARTIFACTS.md` at the pack root).

A recurring task becomes a skill the agent applies by itself: a method, a trigger description and a place in the pipeline — not a prompt that worked once.

## When to use

- The same task is explained for the third time — the repetition is the signal.
- A borrowed method is adapted repeatedly and the team wants its own version.
- An existing skill fires wrongly, misses its triggers, or has no artifact contract — improve it, do not duplicate.
- The pack needs a new step in a route (see `CATALOG.md`) and no existing skill is close enough.
- Not for a one-off prompt (`instruct-agents`) or a document for humans (`cowrite-doc`).

## Inputs

- the recurring task: the runs that repeated, `.orbit/task/<id>/plan.md`, `.orbit/learnings/` for the method that worked;
- `docs/SKILL-STANDARD.md` — required sections, word budget, quality bar;
- `ARTIFACTS.md` — which `.orbit/` or docs path the skill reads and writes;
- `CATALOG.md` and `CATALOG.ru.md` — phase, step number and neighbours;
- the pack: `ls plugins/orbit/skills/` and a search for the trigger phrases before inventing anything.

If an input is missing, stop and get it — a skill without a stable artifact contract becomes an orphan step.

## Procedure

1. **Define the task and triggers.** What the skill does, when it fires (RU and EN phrases), what goes in, what comes out. Check for a duplicate first:

   ```bash
   ls plugins/orbit/skills/
   rg -n "<trigger phrase>" plugins/orbit/skills/*/SKILL.md
   ```

   A similar skill exists → improve it; do not duplicate a capability.
2. **Place it in the Orbit pipeline.** Phase and micro-phase, which skill comes before and after, which `.orbit/` artifact it reads and writes (`ARTIFACTS.md` at the pack root). A skill that writes nowhere is a chat prompt.
3. **Create the folder structure** next to the other skills:

   ```text
   plugins/orbit/skills/<verb-name>/
     SKILL.md            # frontmatter name + description, then the method (English)
     tools/              # scripts, theme, assets — only if the method needs them
   ```

   `name` must equal the folder name and stay kebab-case; keep `[Orbit X.Y.Z Phase]` in `description`.
4. **Write the description.** What it does + when to use + 2–4 trigger phrases in the users' languages (RU and EN); ≤1024 characters; keep the phase prefix. This field decides whether the skill ever fires.
5. **Write the body** per `docs/SKILL-STANDARD.md`: When to use / Inputs / Procedure (6–12 steps with real commands) / Output (artifact path + template) / Check (commands + pass rule) / Pitfalls (≥4) / Portability / What's improved / Based on. Each step names the file or command and what to check.
6. **Validate the structure.** `python3 tools/scripts/quick_validate.py plugins/orbit/skills/<verb-name>` must print "Skill is valid!".
7. **Test the triggers on real requests.** 3–5 prompts the skill should fire on and 2–3 it should not. Tune the description with `tools/scripts/run_eval.py` and `tools/scripts/run_loop.py`; inspect the report via `tools/scripts/generate_report.py` or `tools/eval-viewer/generate_review.py`. Eval definitions: `tools/references/schemas.md` (`evals/evals.json`).
8. **Package if the skill will be distributed.** `python3 tools/scripts/package_skill.py plugins/orbit/skills/<verb-name> [dist]` produces a `.skill` archive; exclude `evals/` and caches.
9. **Wire it into the pack.** Add the row to `CATALOG.md` and `CATALOG.ru.md`; if the skill changes a route, update the router `plugins/orbit/skills/orbit/SKILL.md`. Then run `python3 scripts/validate_pack.py` — it checks 89 skills, names, frontmatter, relative links and junk files.
10. **Ship it with evidence and credit.** Run `python3 scripts/check_skills.py --skill <verb-name>` (must be OK, score ≥ 8.0). Keep the "Based on" section with the source and licence when the method is borrowed; vendor borrowed files under `tools/` with `tools/LICENSE.upstream`.

## Output

A new or improved skill at `plugins/orbit/skills/<verb-name>/SKILL.md` — self-contained, 800–1300 words, with a tested description and a catalog row:

```text
plugins/orbit/skills/changelog-entry/
  SKILL.md          # frontmatter + method
  tools/            # optional: scripts/assets the method cites
```

Record the reason for the skill (the repeated task) and the trigger test results in `.orbit/learnings/`, and mention the new skill in the next `write-update`.

## Check

```bash
python3 tools/scripts/quick_validate.py plugins/orbit/skills/<verb-name>       # Skill is valid!
python3 scripts/check_skills.py --skill <verb-name>                            # OK, score ≥ 8.0
python3 scripts/validate_pack.py                                               # exit 0
rg -c "\[Orbit [0-9]+\.[0-9]+\.[0-9]+ [A-Za-z &]+\]" plugins/orbit/skills/<verb-name>/SKILL.md
```

Pass rule: name matches folder; description has what + when + RU/EN triggers and is ≤1024 characters; body has every section in order, ≥2 code blocks, ≥4 pitfalls, ≥4 commands, ≥2 artifact paths; `quick_validate`, `check_skills` and `validate_pack` exit 0; triggers fire on intended requests and stay quiet on unrelated ones.

## Example

A recurring request — "every merged change needs a changelog entry" — becomes a skill:

```bash
mkdir -p plugins/orbit/skills/changelog-entry
cat > plugins/orbit/skills/changelog-entry/SKILL.md <<'EOF'
---
name: changelog-entry
description: '[Orbit 3.3.5 Docs & writing] Add the changelog entry for a merged change: changelog line, category, link. Use when: «добавь в changelog», «запись в changelog»; changelog entry.'
---
# changelog-entry — …
EOF
python3 plugins/orbit/skills/create-skill/tools/scripts/quick_validate.py plugins/orbit/skills/changelog-entry
# Skill is valid!
python3 scripts/check_skills.py --skill changelog-entry          # OK, score ≥ 8.0
python3 scripts/validate_pack.py                                 # after the CATALOG rows are added
```

Then test it: "добавь запись в changelog для PR #42" and "add a changelog line for the new flag" must fire; "rename this variable" must not. Fix the description until both hold, and record the result.

## Pitfalls

- **Duplicating an existing skill.** Two skills with overlapping triggers fight, and the worse one sometimes wins. Search the pack before creating.
- **No artifact contract.** A skill whose output is "the conversation" cannot be piped into the next step; give it a real `.orbit/` or docs path.
- **A description without triggers, or over 1024 characters.** The body is perfect and the skill never loads; the description is the trigger, not a summary.
- **Skipping `quick_validate` and `check_skills`.** Frontmatter typos and missing sections are cheap to catch and expensive to discover in use.
- **Adding the skill to one catalog only.** `CATALOG.md` and `CATALOG.ru.md` drift; the pack then documents a skill that does not exist or misses one that does.
- **Borrowed tools without a licence.** Vendored files need `tools/LICENSE.upstream` and a "Based on" line; otherwise the pack cannot be shared.
- **A method with invented commands.** Concrete commands are the point; if one cannot be executed, mark it as a placeholder.
- **Description above 1024 characters.** The trigger text gets truncated or rejected; count it.

## Portability

Skills are plain Markdown and work in Claude Code, Codex, Qwen Code, GLM/Z.ai, DeepSeek, Cursor and any agent that reads a skill folder. Validation scripts are Python 3; `quick_validate.py` needs PyYAML, the rest is standard library. Without a skills directory, read `SKILL.md` into context and follow it literally. No slash commands or subagents required.

## What's improved vs the source

- The skill is wired into the pipeline at once: phase, neighbors, `.orbit/` artifacts.
- A mandatory "based on" section with source and licence when the method is borrowed.
- Skill text in English to save tokens; triggers in the users' languages.
- Three gates — structure, trigger eval, pack validation — instead of "looks good to me".

## Based on

- [Skills by Anthropic](https://github.com/anthropics/skills) — method «skill-creator» (Apache-2.0). Rewritten in our own words and improved — see "What's improved".
- Helper files in `tools/` are taken from the source unchanged; licences: `tools/LICENSE.upstream`.
Files in this package (18)
  • SKILL.md 9 KB
  • tools/LICENSE.upstream 11 KB
  • tools/agents/analyzer.md 10 KB
  • tools/agents/comparator.md 7 KB
  • tools/agents/grader.md 9 KB
  • tools/assets/eval_review.html 7 KB
  • tools/eval-viewer/generate_review.py 16 KB
  • tools/eval-viewer/viewer.html 44 KB
  • tools/references/schemas.md 12 KB
  • tools/scripts/__init__.py 1 KB
  • tools/scripts/aggregate_benchmark.py 14 KB
  • tools/scripts/generate_report.py 13 KB
  • tools/scripts/improve_description.py 11 KB
  • tools/scripts/package_skill.py 4 KB
  • tools/scripts/quick_validate.py 4 KB
  • tools/scripts/run_eval.py 11 KB
  • tools/scripts/run_loop.py 13 KB
  • tools/scripts/utils.py 2 KB

Versions

RepositoryScoreLicence
This page · tojim00d/Orbit-skills100MIT
mgiovani/cc-arsenal100MIT
zylos-ai/zylos-core100MIT
mthines/agent-skills100MIT
jongio/skills100MIT
fairladyz625/harness-anything100AGPL-3.0
coco-research/coco100MIT
builderio/agent-native100MIT
BrandeisPatrick/silo-harness100MIT

Database snapshot: 13.11.2026 · package version 6459fdb86fda (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100write-skill
    Create a new agent skill, or improve an existing one, inside this skills repo, following its housekeeping rules. Use when asked to "write a skill", "make this a skill", "save this as a skill", "turn what we just did into a skill", or "improve the X skill".
  • 90skill-writing-ru
    Как писать скиллы и тексты для агентов (SKILL.md, CLAUDE.md, AGENTS.md, брифы субагентам): описания-указатели, ветки, критерии готовности, позитивные формулировки, чистка. Use when: «напиши скилл», «улучши SKILL.md», «почему скилл не срабатывает», «бриф агенту».
  • 100instruct-agents
    [Orbit 3.3.2 Docs & writing] Instructions for agents: CLAUDE.md, prompts, skill descriptions — without contradictions. Use when: «CLAUDE.md», «промпт для агента», «AGENTS.md», «напиши правила для агента», «системный промпт»; agent instructions, system prompt for a coding agent.
  • 90orbit
    Creates, executes and reviews Orbit tasks; searches task history and friction records; records evidence. Use for an assigned task or task ID, task authoring, `orbit.task`/`orbit.search` MCP tools, `orbit tool run`, task records under `.orbit/`, or friction. Use orbit-orchestrate…
  • 100orbit
    [Orbit 0.1.1 Entry] Entry point: sizes the task, proposes a 3–7 step route through the pipeline and walks it. Use when: «с чего начать», «что умеет пак», «проведи задачу от начала до конца», «какой скилл взять», «что делать дальше», «проведи по пайплайну»; where to start, run…
  • 100writing-skills
    Author/rewrite SKILL.md inside .claude-plugin/dinopowers/, enforce wrapper pattern (tea-rags enrichment step then chain into superpowers:*) + description style to win against parent skills. Triggers on "add dinopowers:X", "create a wrapper skill", "write a SKILL.md", "напиши…

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.