Frame Intent

Outcome

Give the next agent or workflow a desired outcome, a set of assumptions, and leading indicators that ground specification and support steering through evidence. The primary deliverable is usable intent. A secondary benefit is helping the human understand why outcome framing, discovery, and evidence matter as they work with agents. Do both through the same work.

Outcome Indicators

• The next agent can specify or design a learning step without inventing the initiative's purpose, critical assumptions, or evidence for its next decision.
• An owner can use an early indicator to decide whether to continue, adjust, investigate, or stop before waiting for the final business result.
• When coaching is needed, the human can explain the relevant outcome, assumption, or indicator choice in their own context. An explanation delivered is not evidence that human behavior changed.

Discovery Question

What intent and evidence should guide the agents before we specify what to build?

Gotchas

• The initiative is never its own goal. "Roll out the copilot" names a solution; find the result someone would still own if that solution were cancelled.
• A polished brief does not make a belief true. Preserve unknowns rather than inventing baselines, targets, owners, or supporting evidence.
• An outcome is a change in what someone does. Tool usage and things shipped can miss the actual change in work; connect that change to a business result.
• A metric is leading relative to an outcome and a decision. State the causal belief and the action that different evidence could prompt.
• Coaching must not become a prerequisite course. Draft what context supports; ask only about gaps that change the next decision.

Intake and working style

Read the supplied request, existing artifacts, and constraints first. Establish the unit of work and the receiving workflow's native intake or spec format. This skill works in conversation or from supplied material without requiring a particular framework, tool, or web access.

With enough context, synthesize the brief directly and invite correction where it matters. With thin context, ask one or two consequential questions at a time and update the brief as answers arrive. Separate confirmed inputs, inferences, proposals, and unknowns. If no human is available, produce a provisional brief and flag decisions that still need an owner; never invent approval or block useful drafting on an interview.

Coach when a choice needs explanation: briefly explain why a proposed goal is an output, why an assumption needs discovery, or how an indicator changes a decision. Use the person's example and invite correction. Adapt to familiarity and requests for brevity; do not repeat understood lessons or add a mandatory quiz. The artifact still has to be usable.

Frame the intent

1. Find the owned result. State the desired outcome, accountable owner, why it matters now, and baseline, target, and time horizon where known. For "maintain" or "protect," establish today's result. Separate the buyer, the people doing the work, and the person whose numbers move.
2. Identify the constraint and boundaries. Name what plausibly holds the result back, evidence for that belief, non-goals, and hard constraints. Would fixing it require the proposed solution at all? Keep solutions as options. For mandatory compliance or safety work, state the fixed obligation honestly and investigate uncertainty about effectiveness or adoption where it exists.
3. Name the change in work. Which people or agents will do what differently compared with today, and how should that affect the result? For internal or technical work, include the downstream beneficiary. "People use the tool" is not enough; consider adoption alongside delivery.
4. Expose assumptions. State a falsifiable causal belief: if this change in work happens, this result should follow, because of this mechanism. Record critical assumptions, evidence for and against, confidence with a reason, and observations that would disprove them. Use stable IDs for handoff. Prioritize assumptions whose failure changes the bet; use desirability, usability, feasibility, and viability as prompts where useful.
5. Choose leading indicators for steering. Select a small set of early signals that could change the next decision. For each, give the linked outcome or assumption, why it should lead it, measure and population, baseline, proposed target or threshold, source, owner, review timing, and action different results would prompt. Mark missing values; collecting a baseline can be the next step. Distinguish early behavioral signals from final results and technical acceptance tests. Demos or completed work do not establish impact.
6. Choose the next learning or delivery step. With weak evidence for a critical assumption, recommend the smallest useful experiment: what to learn, how, by when, what changes the decision, and who decides. With sufficient evidence, hand intent to specification with remaining assumptions visible. Do not impose discovery when proportionate evidence already supports delivery. Record the owner and trigger for revisiting intent after release and adoption; shipping alone cannot establish that the outcome occurred.

Output and handoff

Produce a concise intent brief with these contents, adapting headings to the receiving framework rather than imposing a template:

• Desired outcome: owner, who does what differently, expected business result, baseline, target, and time horizon; label missing or proposed values.
• Context and boundaries: why now, constraint and evidence, scope, non-goals, fixed obligations, and solution options kept separate.
• Assumptions and evidence: causal hypothesis, critical assumptions, supporting and opposing evidence, confidence, and falsification signals.
• Leading indicators and steering: measures and causal links, baselines, targets or thresholds, sources, owners, review cadence, and implications for continuing, adjusting, investigating, or stopping.
• Next decision: discovery or specification, next learning step where needed, decision owner, timing, unresolved questions, and outcome-review trigger.
• Status of claims: confirmed, inferred, proposed, and unknown separated.

Use the caller's existing artifact or native spec section as the canonical home. If a standalone file is needed, use an intent-named Markdown file in the actual workspace, such as intent.md, and link it from the receiving spec. In chat-only settings, return the brief in chat. Do not create a competing copy or change a receiving framework's schema or instructions without authorization.

Tell the receiving agent to preserve the outcome, assumption IDs, and indicator links when deriving scenarios, requirements, or an experiment. Architecture and stack choices belong to planning unless supplied as hard constraints. Flag gaps that prevent a particular decision without discarding the usable brief. Human approvals remain human.

Boundaries

Use this at any scale: a product opportunity, feature, experiment, internal capability, operational change, or portfolio investment. Route by the question and next decision, not by the work item's label or the framework's vocabulary.

• Own the intent and its evidence needs. Clarify the desired outcome, assumptions, leading indicators, and next decision. Recommend discovery where evidence is weak, but do not imply that framing has tested an assumption, completed discovery, or authorized a commitment.
• Keep the next steps linked. Discovery tests assumptions; specification derives requirements and acceptance scenarios; planning chooses an approach. Hand them the intent and unresolved questions without taking over those steps or imposing a different framework, artifact, or stage sequence.
• Use lighter help when intent is already established. If only wording needs work, reframe the statement without restarting the full process. outcome-framing-coach is one option when available.
• Separate framing from assessment and portfolio steering. Assessing whether evidence supports investment is a readiness question; comparing several bets is a portfolio question. sniff-test and sniff-test-portfolio can support those questions when available. Neither the size of the work nor the presence of an existing document is a reason to skip intent framing when it is needed.

Human investment, priority, and approval decisions stay with their owners.

For maintenance or evaluation, read references/tests.md for trigger cases and behavioral acceptance scenarios.

References/Inspiration/Credits

The inspiration for this skill is Yuval's work introducing intent framing into portfolio/product development life cycles and coaching product professionals to shift from specifying requirements to framing intent.

To dive deeper into examples of intent framing in modern product management, check out the Lean Product Canvas by Jeff Gothelf and Josh Seiden, and Mark Richards' feature canvas. These illustrate ways to frame intent; this skill does not require or reproduce either artifact. "Who does what by how much" is Gothelf and Seiden's phrasing. See CREDITS.md for the wider lineage.

These are Yuval's methods, not his personal assessment of a particular initiative.