agentsclimarketplace
Sign in

Write agent

Skill tale-project/tale/.agents/skills/write-agent

Use this skill whenever you create or update an agent in a Tale organization — a chat assistant, an image generator, or an external coding agent (Claude Code / Cursor). Triggers include: "create an agent", "change an agent's instructions, model, or tools", "give an agent a skill or integration", or an automation that needs its own agent. Never author an agent definition without it — and never create one before listing the existing agents and ruling out reuse. For the trigger→steps definitions an agent participates in, use write-workflow; for the installable bundle that groups agents, use write-automation.From its SKILL.md

Install
npx -y skills add tale-project/tale --skill write-agent

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 17 stars17 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

4.9 KB, ~1.1k tokens by cl100k_base, as published. Nobody here has run it

write-agent

An agent is one JSON definition. Org-global agents live in the org config at agents/[folder/]<name>.json — the folder is organizational only, never identity. An automation's own agents live inside its bundle under apps/<automation>/agents/ (see write-automation) and are addressed as <automation-slug>/<agent-slug>. The definition must parse against the platform's agent schema; an invalid file is rejected at save with the failing field named.

Reuse before you create — tick every box

  • List the existing agents of the org (roster + catalog) first.
  • One already solves the request fully → use it; create nothing.
  • One solves it partially → update it, or use it as the base and extend.
  • Nothing fits → only then author a new definition.

A divergent near-duplicate of an existing agent is a defect, not a feature.

Identity and naming

  • slug is the canonical identity, stored in the file — flat lowercase a-z0-9_-, must start alphanumeric, ≤64 chars, no / (^[a-z0-9][a-z0-9_-]*$). Moving or renaming the file never changes identity.
  • An automation-owned agent's effective name is the composite <automation-slug>/<slug> — at most one /, and only for automation-owned agents.
  • Never reuse a reserved slug (auto, organigram) or overwrite a protected system agent (assistant, workflow-assistant).

Required fields

  • displayName (1–200 chars) — top level or in at least one i18n locale.
  • systemInstructions (≤20 000 chars) — required for chat agents (the default kind), top level or in at least one locale.
  • supportedModels — at least one model ref in [provider:]model-id form. Only a BYO external agent may leave it empty.

The fields that matter

  • description (≤1000) and conversationStarters (≤4 entries, ≤200 chars each).
  • skillBindings — kebab-case skill slugs, max 10. This is the agent's whole skill set: empty or absent means the agent has NO skills; there is no implicit "all org skills".
  • toolNames — platform tools the agent may call; integrationBindings — integration slugs it may operate; workflows — workflow slugs it may run.
  • primaryBehavior — chat (default) | image-generation | external-agent. The non-chat kinds bypass the platform tool loop, so they reject toolNames/workflows (external-agent keeps integrationBindings as its sandbox grant set). agentKind (claude-code | cursor) and authMode (managed | byo) are external-agent-only; Cursor must be byo.
  • knowledgeMode / webSearchMode — off | tool | context | both.
  • Limits: maxSteps (1–100), timeoutMs (≥1000), maxConcurrentTasks (1–50), budget (monthlyCents plus warnPct ≤ pausePct).
  • roleRestriction: "admin_developer" gates the agent to admins/developers; visibleInChat shows it in the chat picker; routing.modelSelection config | auto picks the model per turn.
  • i18n — per-locale overrides of displayName, description, conversationStarters, systemInstructions, keyed by locale (en, de, fr-FR, …).
  • metadata — autoInstall (seed enabled into every org), templateCatalog, labels (≤12; the first is the catalog section), requires.integrations (hard dependency — the agent is disabled while any listed integration is unconnected), requires.env (keys the install wizard collects).

Minimal valid example

{
  "slug": "faq-helper",
  "displayName": "FAQ Helper",
  "systemInstructions": "You answer questions about the company FAQ, concisely.",
  "supportedModels": ["openrouter:anthropic/claude-sonnet-4.6"],
  "skillBindings": ["web-research"],
  "visibleInChat": true
}

Validation

Every save parses the file against the shared agent schema — a violation is rejected with the exact field path and message (e.g. supportedModels: At least one model is required.). Keep the definition minimal: omit optional fields instead of guessing values.

Siblings: write-workflow (the processes agents act in), write-skill (what skillBindings points at), write-integration (what integrationBindings points at), write-automation (bundling agents into an installable automation).

Gives 0 of the 12 instructions most docs writing skills give in ~1.1k tokens

Counted across 1,637 of the 3,044 authors here whose files we hold, read 2026-08-07

  • Announce the skill at startin 54 of 1637, across 26 files
  • Convert legacy doc files before editingin 45 of 1637, across 7 files
  • Predict questions readers might askin 42 of 1637, across 4 files
  • Generate clarifying questions for initial contextin 42 of 1637, across 3 files
  • Create document scaffold with placeholder textin 42 of 1637, across 3 files
  • Brainstorm content options for each sectionin 42 of 1637, across 3 files
  • Test the document with a fresh context-less instancein 42 of 1637, across 3 files
  • Include exact file paths in every taskin 42 of 1637, across 15 files
  • Ask interview questions one at a timein 42 of 1637, across 27 files
  • Apply surgical edits during refinementin 41 of 1637, across 2 files
  • Offer structured workflow or freeformin 40 of 1637, across 1 file
  • Ask for document meta-contextin 40 of 1637, across 2 files

Said here and by no other author read

  • author agent definitions as JSON files
  • ensure definitions parse against the agent schema
  • use lowercase slug as canonical identity
  • include required displayName and systemInstructions fields
  • include at least one supported model reference
  • omit optional fields instead of guessing values

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 326,851. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.