Release v2.5.3 — close the benchmark learning loop (brain benchmark-credit)#38
Merged
Conversation
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
audit-check@v2 reports warnings via a check run, which needs checks:write on GITHUB_TOKEN — the top-level contents:read block was 403ing the reporting step even though the audit passed. Also updates cxx (transitive via usearch) past the let_cxx_string unsoundness advisory published today. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Diagnosis (docs/superpowers/tb2-learning-loop-diagnosis.md): the TB2 learning loop measured nothing across two iterations because the benchmark cite path (MCP kimetsu_brain_cite -> record_mcp_citation) is a SINGLETON dead-end — one id, no query, fresh run — that feeds none of usefulness/staples/routes. Plus Harbor collapsed the codex MCP command+args so the agent ran blind (fixed host-side). New: kimetsu brain benchmark-credit --task <t> --passed. On a pass it grouped-cites the top-k memories most relevant to the task via the record_citations path (grouped + query) — the same path consolidation v1 consumes. Driven host-side by the harness after grading, so the learning signal never depends on the flaky in-container agent. Verified: 6 routes built from 2 credited tasks; container agent now calls kimetsu tools after the harbor config fix. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
# Conflicts: # CHANGELOG.md # Cargo.toml # crates/kimetsu-agent/Cargo.toml # crates/kimetsu-brain/Cargo.toml # crates/kimetsu-chat/Cargo.toml # crates/kimetsu-cli/Cargo.toml # crates/kimetsu-e2e/Cargo.toml
…ln ref - crossbeam-epoch 0.9.18 -> 0.9.20 clears RUSTSEC-2026-0204 (invalid pointer deref in the fmt::Pointer impl). - drop a redundant & in a prune-display println! that newer stable clippy flags with -D warnings (brain.rs:2760). Co-Authored-By: Claude Opus 4.8 (1M context) <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
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.
Focused follow-up to v2.5.2.
What ships
kimetsu brain benchmark-credit --task <t> --passed: on a graded task pass, groups a citation for the memories most relevant to the task and routes it through the same grouped, query-linked path consolidation consumes (usefulness, query-routes, staples). Driven host-side by a benchmark harness after grading, so the learning signal never depends on the agent inside the sandbox calling a tool.Gate
Release build clean, clippy
-D warningsclean. benchmark-credit itself already carried its brain tests in v2.5.2's suite.After merge
Tag
v2.5.3to fire the release workflow (binaries + npm with provenance).🤖 Generated with Claude Code