Rewrite generate-theories skill UX after dogfooding#62
Merged
Conversation
The skill was previously a thin wrapper around the Theorizer A2A CLI. It now choreographs the full request end-to-end. - Two-half structure (reference / flow) separated by a "How to run it" banner and orientation paragraph. - Question shaping: four-dimension rubric (Intervention, Outcome, optional Mechanism, Scope) with explicit audit step and two routes — proceed silently when all required dimensions are present, or poll for what's missing. - Escape hatch in Route 2: "just send it," "let Theorizer figure it out," etc. bypass shaping for advanced users who want to under-specify deliberately and let Theorizer infer from papers. Surfaced in the poll example. - Refuse path for non-research questions explains Theorizer's domain instead of forcing the rubric. - Mode selection (automatic vs piecemeal) via AskUserQuestion, with Continue / Edit / Stop checkpoints between piecemeal stages. - Pipeline section refreshed against the live agent card: adds build-extraction-schema, tightens stage descriptions, reflects form-theory / evaluate-novelty accepting either a prior task_id or user-supplied data. - Internal scaffolding stays internal — explicit rule that audit results, dimension names, and route numbers don't appear in user-facing chat. - Diverse non-AI-biased well-formed examples (metabolic medicine, education, ecology) to avoid biasing generations toward chatbot scopes. - Fix Theorizer GitHub URL → asta-theorizer-internal. - Add evals/evals.json with three baseline prompts spanning the silent / poll / refuse routes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
make build-plugins, picking up the generate-theories SKILL.md rewrite and the new evals/evals.json. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Validator requires a space-separated string, not a YAML list. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Will live elsewhere per review feedback. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
"just send it" → "skip that" in the rule list. Poll example rewritten in
the agent's voice ("I can send your query verbatim...") rather than
echoing a quoted user phrase.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
charliemcgrady
approved these changes
May 21, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Updated the
generate-theoriesskill UX. The skill now choreographs the full request end-to-end via question shaping, mode selection, stage checkpoints, and handoff.What changed
AskUserQuestion, with Continue / Edit / Stop checkpoints between piecemeal stages and decision-oriented per-stage summaries.Newevals/evals.json— test prompts spanning the silent / poll / refuse routes for future iteration.