feat(debate): confidence-gated multi-agent iterative debate for hard sorries - #119
Open
romirthedev wants to merge 2 commits into
Open
feat(debate): confidence-gated multi-agent iterative debate for hard sorries#119romirthedev wants to merge 2 commits into
romirthedev wants to merge 2 commits into
Conversation
Before each reply in the /learn iterate loop, the skill now reasons from three advisor perspectives (Pace, Method, Depth) to select the best response strategy for the learner at that moment, rather than immediately continuing. The chosen strategy is surfaced as a brief *Pedagogy: ...* note before the actual reply. The debate is mandatory in game and socratic modes, optional in tour and exercise modes. New behaviors: - Hint escalation ladder in game mode: directional hint → specific hint → full answer, triggered after 2 consecutive failures on the same exercise - Stuck detection: if the learner shows the same misunderstanding twice, the strategy must switch framing — never repeat the same explanation - Mid-session profile updates: debate can raise/lower --level or switch --style based on observed learner signals, announced inline Updated files: - commands/learn.md: new step 4.5 between Depth Check and Iterate; updated step 5 to reference it - references/learn-pathways.md: new ## Pedagogical Self-Debate section with full advisor table, hint escalation protocol, stuck detection spec, and profile update rules
…ard sorries - Add 4 independent debate agents: Mathematician, Tactician, Skeptic, Judge - Agents debate iteratively with full shared history until convergence - Confidence gate: single agent self-assesses 100% confidence before spawning debate - Judge detects convergence each round; hard stop at max_rounds (default 5) - Wire debate loop as imperative steps in cycle-engine.md so it runs automatically - Add --debate and --debate-max-rounds flags to prove and autoprove - Difficulty rubric (1-10) gates the confidence check at score >= 7 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
32 tasks
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
1 participant
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.
Summary
cycle-engine.mdso it runs automatically during/lean4:proveand/lean4:autoprove— no manual chaining needed.--debateand--debate-max-roundsflags toproveandautoprove.Test plan
claude --plugin-dir ./plugins/lean4/lean4:prove --debate=askon a hard sorry (e.g. measure-theoretic goal withℝ≥0∞) and verify debate triggers automatically--debate=offdisables agent spawning while still logging difficulty scoreexecution_plan🤖 Generated with Claude Code