Behavior Design Agent Skill

Behavior Design

<skill_contract>

<input>Actor, desired outcome, current behavior, context, constraints, baseline evidence, and review horizon.</input>

<output>An agency-preserving HAS behavior system with cues, effort levels, SOP, measures, recovery, and review cadence.</output>

<done>The target behavior is observable, runnable on a low-motivation day, evidenced, and assigned one review owner.</done>

<non_goals>Coercion, unsupported identity claims, vague motivation advice, or optimizing lagging outcomes without behaviors.</non_goals>

Convert an outcome into the smallest observable behavior system that can survive low-motivation days. Preserve agency; optimize the environment before blaming the actor.

Usage Template

Provide: desired outcome, actor, context, current behavior, constraints, review horizon, and available evidence. Optional: failed attempts and environmental cues.

Workflow

<intake>

• Restate the outcome as an observable change, not an identity label.
• Establish baseline frequency, friction, trigger context, and hard constraints.
• Separate controllable behavior from lagging outcome.
• Set one review horizon and one owner.

</intake>

<unknowns_gate>

Classify unknowns as known, probeable, testable, or blocked. If actor, target behavior, or safety boundary is absent, return NEEDS_INPUT with one minimal question. A reversible assumption is allowed only when labeled and paired with a same-cycle test.

</unknowns_gate>

<execute>

Build the HAS sequence:

• H1 Goal: translate the outcome into one leading behavior and one lagging measure.
• H2 Habit: define anchor, cue, location, and three effort levels: minimum (about 2 minutes), normal, and stretch.
• H3 SOP: write trigger -> action -> evidence -> recovery -> stop so a missed run has a next move.
• H4 Review: select a cadence; compare planned versus observed behavior and change one variable only.
• H5 Identity: use evidence-based reinforcement: “I am becoming X because I repeatedly did Y,” never unsupported affirmation.

Reduce friction before adding motivation. Prefer defaults, visible cues, prepared tools, and short feedback latency. Keep the minimum behavior useful rather than ceremonial.

</execute>

<evaluate>

Check that the behavior is observable, starts in a named context, fits the minimum effort budget, produces evidence, and has a recovery path. Reject plans that depend on constant willpower, hide coercion, or measure only outcomes. If the check fails, revise the highest-friction element once and re-evaluate.

</evaluate>

<state_contract>

Persist {run_id, status, attempt, budget, evidence, unknowns, last_error, next_action} plus baseline, current H1-H5 design, review date, and revision history. Update atomically after each review; never overwrite prior observations.

</state_contract>

Failure Protocol

• NEEDS_INPUT: a target, actor, or boundary is missing; ask one narrow probe.
• INSUFFICIENT_EVIDENCE: baseline is unknown; run a short observation period before optimizing.
• VERIFY_FAILED: the behavior is not observable, feasible, or linked to the outcome; revise one variable.
• BUDGET_STOP: the review horizon or effort budget is exhausted; preserve state and report the next experiment.

Output Contract

Return status, result (H1-H5 plan), evidence (baseline and measures), unknowns, and next_action (owner plus review date).

Edge Cases

• Motivation is near zero: keep the cue and reduce the action to a useful minimum; do not expand the reward system.
• Repeated misses occur despite compliance: treat the goal-behavior link as unverified and test a different leading behavior.

Success Metrics

• One named cue reliably starts one observable minimum behavior.
• Execution evidence is captured at the chosen cadence.
• Reviews change the system using observed friction, not self-judgment.

Quality Gates

• [ ] Outcome, leading behavior, and lagging measure are distinct.
• [ ] Minimum action, recovery path, owner, and review date exist.
• [ ] The design preserves agency and has no coercive mechanism.
• [ ] Completion claims use observed behavior evidence.

</skill_contract>