mcpField
A
agently-output-control

agently-output-control

github.com/agenteraAgent workflows3141 installsUpdated 4d ago
Repository

About agently-output-control

A skill for controlling the shape and reliability of Agently model outputs. It guides choices among request-level contracts, `.output(...)`, and `ensure_keys`, including async result handling.

Key features

  1. 01Choose output contract placement
  2. 02Use `.output(...)` for dynamic schemas
  3. 03Enforce required fields with `ensure_keys`
  4. 04Prefer async handling when appropriate

Use cases

  1. 01Return stable structured reports
  2. 02Provide machine-readable fields to downstream steps

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/agentera/agently-skills --skill agently-output-control

After installing, type/agently-output-controlin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

examples/
.gitkeep
minimal.py
outputs/
.gitkeep
expected.md
references/
overview.md
scripts/
check.py

Details

From the project README

Agently Output Control

Use this skill when the question is what shape the model should return and how that shape should stay reliable.

The user does not need to say .output(...) or ensure_keys. Requests for stable JSON-like fields, structured reports, or machine-readable sections should route here.

Native-First Rules

  • default to async-first response consumption when structured output will be streamed, reused, or served over an async boundary
  • prefer prompt-config-owned output contracts such as .request.output when the schema is stable and shared across a request family
  • prefer .output(...) for machine-readable results when the schema is dynamic, exploratory, or easier to keep close to code
  • prefer ensure_keys when required fields must be enforced
  • keep output schema explicit when downstream systems, workflow branches, or later model steps consume the result

Anti-Patterns

  • do not handwrite JSON post-processors when .output(...) already owns the contract
  • do not rebuild a stable shared schema in Python if prompt config can own it once
  • do not build custom retry loops for missing keys before checking ensure_keys
  • do not default to sync-only result handling when the caller is already async-capable

Read Next

  • references/overview.md

Comments

Markdown supported