# Reframe

> Use when asked to reframe a task, raise its ambition, improve a prompt, or define acceptance criteria, or when a workflow needs a brief.

- Skill: `sreejithraman/reframe` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx get-skill add sreejithraman/reframe`
- Raw SKILL.md: https://api.skillmd.com/api/skills/sreejithraman/reframe/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: sreejithraman (https://skillmd.com/u/sreejithraman)
- Updated: 2026-11-16
- Page: https://skillmd.com/skills/sreejithraman/reframe

---


# Reframe

Form a brief worth executing from the user's request and context. Preserve
explicit requirements while considering stronger ways to achieve the outcome.
Scale the pass to what is unsettled: a precise request may need only a concise
confirmation. Reframing alone ends with the brief.

## Find the better task

Read the request, settled decisions, and relevant project evidence. Separate
the desired result from the proposed means or artifact. Keep alternatives beyond
authorized scope labeled as proposals. Use the
[acceptance guidance](references/acceptance.md) to resolve
what context can establish and ask for material user decisions.

Consider the dimensions that can improve this task:

- **Outcome:** who needs what change, and what decision or activity does the
  requested artifact serve? When useful, compare the artifact, the outcome it
  serves, and a lasting workflow that improves that outcome. Explore whether a
  substantially harder target is feasible; distinguish proposed stretch targets
  from the agreed finish line.
- **Depth:** which original inputs, expert methods, and instruments would make
  the work more rigorous? Test consequential assumptions against raw evidence.
  Consider what new finding could change the decision and how to test it;
  predictive claims need evaluation on unseen evidence. Name the method rather
  than an expert persona.
- **Range:** describe what the work must accomplish as activities, then ask which
  adjacent field handles them well. Choose a concrete technique and its tools.
- **Evaluation:** how and by whom should the result be judged, which constraints
  must hold, and what finish line was requested? Use the acceptance guidance to
  propose suitable checks or qualitative criteria, leaving uncertainty and
  missing prerequisites explicit.

Choose the strongest coherent framing rather than expanding every dimension.
Raise the outcome or standard; extra deliverables must earn their place.
Use [Research](../research/SKILL.md) for factual uncertainty worth investigating.
Use experiments to settle empirical questions within the requested scope.
Keep ambitious targets distinct from evidence that they have been achieved.

Name the data, tools, environments, and access needed to execute and evaluate
the chosen approach, distinguishing available resources from missing setup.
For a lasting workflow, identify its signal, action, feedback, and owner. Keep
setup and ongoing operation within the authorized scope; a proposal alone
authorizes neither automation nor recurring execution.

## Return the brief

State the outcome and target, the code it covers, preserved constraints, what
is out of scope, relevant methods, evaluation, and requested finish line. Include material assumptions and unknowns;
explain meaningful framing changes. Proposed evaluation is not evidence that
the result has been achieved. Reuse the existing format and task record; a clear
request can be confirmed in a few sentences without a separate document.
When prompt rewriting is the requested product, return a copyable prompt with
the outcome, preserved constraints, methods, and evaluation.

When another workflow calls Reframe, return the brief to that owner. The owner
continues execution; do not start another copy of its workflow.

For a direct reframing request, return the brief in chat unless another format
or destination was requested. If the user also requested execution, continue
through the authorized workflow with that brief. Use
[Implement](../implement/SKILL.md) for one outcome, as one PR or a planned stack,
[Coordinator](../coordinator/SKILL.md) for several outcomes, and
[Swarm](../swarm/SKILL.md) for a bounded parallel assignment when useful.
Reframing alone authorizes no implementation, native goal,
schedule, or external action.

