Research desk
Source ledgers, claim checks, comparison matrices and freshness scans. For work where a good-looking answer with a bad source is worse than no answer.
Skill library
Skill packs that keep the workflow, not the chat
Six small Agent Skills packs for research, client work, writing, development, agent operations and sales — each with triggers, output contracts, install steps and tests.
6packs
24skills
1open folder format
Client room
Call-to-scope, weekly updates, decision logs and handoff packs. Keeps the agreement visible while the work is moving.
Writing room
Reader briefs, voice matching, claim checks and repurposing. One draft, several formats, no invented facts.
Dev loop
Reproductions, diff reviews, release notes and dependency checks. The checks that make an agent's patch reviewable.
Agent ops
Project briefs, workflow audits, prompt-to-skill conversions and fallback plans. The layer that makes an agent setup survive a new session.
Sales room
Qualification, proposal follow-up, an objection log and a delivery handoff. Keeps the promise visible from first reply to kickoff.
What a skill is
A skill is a folder of instructions, references and optional scripts that an agent loads when a task matches. The open format is small: a SKILL.md file with a name and a description, plus whatever the agent needs to do the job. The description is the trigger; the body is the procedure.
That is different from a prompt. A prompt is a message you send; a skill is a file the agent can find again next month. A pack is a small set of skills that share a job, so you install the workflow once instead of remembering four prompts.
Why packs, not a marketplace dump
Installing every skill you find makes the agent slower and less predictable: more descriptions to match, more instructions to conflict. A pack should be small enough to read in one sitting and specific enough that you can test it.
These six packs cover the work this site is about: research, client work, writing, development, agent operations and sales. They are starting points, not endorsements of any vendor. Adapt the triggers to your own vocabulary and delete the skills you do not use.
The three-part test
A skill earns its place if it passes three checks:
- Trigger — a real request makes the agent load it without you naming the skill.
- Output — the result has a shape you can check, not just a tone you like.
- Guardrail — there is a failure the skill prevents, and you can describe it.
If a skill cannot pass the third check, it is a style preference. Keep it in a prompt, not in a folder.
Security and portability
Skills are instructions, not a security boundary. Read the SKILL.md before you install it, keep secrets out of the folder, and treat any script as code you are about to run. Prefer skills that use the tools you already have and do not ask for network access they do not need.
The folder format is portable across agents that support it. Keep the content vendor-neutral: no model names in the instructions, no tool-specific syntax unless the skill is explicitly for that tool.
Sources
- Agent Skills specification — the folder format and
SKILL.mdfrontmatter. - Anthropic's public skills repository — reference implementations to read before writing your own.
- Claude Agent Skills overview — how skills load and where they run.
For agents
The pack pages are at /skills/<pack>.md; each includes the same skill table, install steps and tests. If a pack assumes a tool or file layout that does not exist in your environment, send the mismatch to the help desk instead of guessing.
Found a gap in the library?
Send the scenario, the prompt that failed, or the harness change that broke your setup. The help desk takes questions and corrections from humans and agents, and every accepted fix updates the page date.