Adopt action-first neurodivergence-friendly output as baseline context #609

New issue

Closed

opened 2026-07-20 15:37:58 +00:00 by coilysiren · 0 comments

coilysiren commented

2026-07-20 15:37:58 +00:00

Owner

Source

https://github.com/ayghri/i-have-adhd

The upstream MIT-licensed package shapes agent output for an ADHD reader. It leads with action, numbers bounded steps, restates state, suppresses tangents, makes completed work visible, uses matter-of-fact errors, and removes filler preambles and closers.

Placement decision

This is baseline person context, not an ordinary skill and not a role-composed skill. The communication shape applies across roles, repositories, harnesses, and task types. Selective discovery would make the preference disappear exactly when it is needed.

Do not install or vendor the upstream package as a discoverable skill. Adapt the useful rules into the canonical baseline source consumed by agent-compose.

Adaptation work

Reconcile the upstream rules with the existing voice, safety, tool-announcement, and name-the-actor conventions.
Preserve action-first answers, bounded numbered steps, visible state, tangent suppression, concrete next actions, and matter-of-fact errors.
Keep explicit exceptions for explanation requests, destructive actions, real ambiguity, debug spirals, and harness requirements.
Treat time estimates as estimates with named uncertainty rather than false precision.
Preserve enough warmth and context for collaborative conversation instead of copying the upstream no-preamble rule mechanically.
Record the upstream license and attribution for any adapted text.

Taxonomy evidence

This issue is a counterexample for the scoped-preference type study: a universally relevant operator preference belongs in baseline person context rather than a skill.

References:

Acceptance criteria

Every composed role and supported harness receives the adapted baseline.
No SKILL.md or COMPOSED.md entry is created for this behavior.
Existing higher-priority safety and harness rules remain intact.
Positive and near-miss prompts verify actionability without forcing every answer into the same length or format.
The implementation identifies and edits the canonical source rather than generated output.

Completing this issue provides taxonomy evidence but does not close the comprehensive taxonomy epic.

## Source https://github.com/ayghri/i-have-adhd The upstream MIT-licensed package shapes agent output for an ADHD reader. It leads with action, numbers bounded steps, restates state, suppresses tangents, makes completed work visible, uses matter-of-fact errors, and removes filler preambles and closers. ## Placement decision This is baseline person context, not an ordinary skill and not a role-composed skill. The communication shape applies across roles, repositories, harnesses, and task types. Selective discovery would make the preference disappear exactly when it is needed. Do not install or vendor the upstream package as a discoverable skill. Adapt the useful rules into the canonical baseline source consumed by agent-compose. ## Adaptation work * Reconcile the upstream rules with the existing voice, safety, tool-announcement, and name-the-actor conventions. * Preserve action-first answers, bounded numbered steps, visible state, tangent suppression, concrete next actions, and matter-of-fact errors. * Keep explicit exceptions for explanation requests, destructive actions, real ambiguity, debug spirals, and harness requirements. * Treat time estimates as estimates with named uncertainty rather than false precision. * Preserve enough warmth and context for collaborative conversation instead of copying the upstream no-preamble rule mechanically. * Record the upstream license and attribution for any adapted text. ## Taxonomy evidence This issue is a counterexample for the scoped-preference type study: a universally relevant operator preference belongs in baseline person context rather than a skill. References: * https://forgejo.coilysiren.me/coilysiren/inbox/issues/249 * https://forgejo.coilysiren.me/coilysiren/inbox/issues/256 * https://forgejo.coilysiren.me/coilysiren/inbox/issues/257 ## Acceptance criteria * Every composed role and supported harness receives the adapted baseline. * No SKILL.md or COMPOSED.md entry is created for this behavior. * Existing higher-priority safety and harness rules remain intact. * Positive and near-miss prompts verify actionability without forcing every answer into the same length or format. * The implementation identifies and edits the canonical source rather than generated output. Completing this issue provides taxonomy evidence but does not close the comprehensive taxonomy epic.