feat(harness): lessons ledger, project-verify gate, quantitative goals, loop docs (v0.6.5) - #24
Merged
Merged
Conversation
…s, loop docs (v0.6.5) Adopts four ideas from Anthropic's "Getting started with loops" article (claude.com/blog/getting-started-with-loops): - Plan files gain a "Lessons for the Guildhall" section; Step 6 has Mordain record gate breaks / retries / misjudged triggers and promote lessons that recur across plan files into the chronicle with a suggested amendment — individual failures become system improvements. - Step 5 GREEN gate extends past the unit suite: if the target repo documents a verification path (verify skill, make verify, lint or typecheck named in CLAUDE.md), Mordain runs it; failures draw from Bruga's existing single-retry budget. - Bruga's dispatch must state the exit criterion quantitatively from the RED-gate counts, making the goal machine-checkable at the gate. - README gains "Quests on a schedule" — composing /quest with /loop and /schedule, with stop-condition and interval guidance. No new command; existing primitives compose. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Adopts four ideas from Anthropic's "Getting started with loops" article (also published via @ClaudeDevs on X). Guildhall already embodies most of the article — second-agent reviews, model-to-complexity tiering, evaluator-separate-from-actor gates — so these are incremental additions, not architecture changes.
quest.mdplan template + Step 6) — the article's "encode individual failures into system improvements." Each quest's plan file records gate breaks, retries, misjudged gating triggers, and blocked adventurers; Mordain greps prior plan files and promotes lessons that recurred across 2+ quests into the chronicle with a concrete suggested amendment. An explicit "none — quest ran clean" clause prevents literal-reader invention.quest.mdStep 5) — after GREEN, Mordain runs any verification path the target repo documents (verify skill,make verify, lint/typecheck named in CLAUDE.md). Failures draw from Bruga's existing single-retry budget so the "ONE retry max" stop condition holds.quest.mdMordain's brief) — the=== EXPECTED OUTPUT ===block must state the exit criterion numerically from RED-gate counts, making the goal machine-checkable at the Step 5 gate.plugin/README.md) — documents composing/questwith/loopand/schedule, with stop-condition and interval guidance. Deliberately docs-only: no new command, no scope creep.Also bumps the plugin to v0.6.5 and updates the CLAUDE.md version reference.
Test plan
scripts/validate_plugin.py— 18 agents checked: 0 errors, 0 warnings## Lessons for the Guildhall, the GREEN gate looks for a project verification path, and Bruga's dispatch states a numeric exit criterion🤖 Generated with Claude Code