Repository navigation
release: 0.27.0 - #461
Merged
Merged
release: 0.27.0#461
Conversation
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Release 0.27.0. This bumps every version surface and turns
[Unreleased]into the dated## [0.27.0]section, which the release workflow extracts as the GitHub Release notes.Why 0.27.0 and not 0.26.1: the CHANGELOG says the project follows SemVer, and the 33+ commits since 0.26.0 include additions: the EXPERIMENTAL
codecarto_changetool (tools/list 22 → 23), the engineering-record core exports, thetraverseskill, and new templates. All of that warrants a minor bump.What ships
2026-07-28(MCP server on SDK v2, serving 2026-07-28 alongside the 2025 era (#185) #443), the EXPERIMENTAL engineering record surface (E01–E11), the Windows fixes (atomicWriteFile: rename over an existing file fails with EPERM on Windows under concurrent writers #393, Pi write guard refuses a legitimate .codecarto/ write on Windows #394, fix: make stale lock ticket reaping single-winner on Windows #444), and the test-isolation fixes.Version surfaces
package.json,package-lock.json(both root fields),server.json(both fields), and.codecarto/workflow/scaffold-version.yamlare all 0.27.0.Candidate verification (head aa19384)
npm cisucceeds andnpm audit --omit=dev --audit-level=highfinds 0 vulnerabilities.npm run buildsucceeds, andnpm testpasses 1401/1401.codecartographer-pi-0.27.0.tgzhas 247 files, 570,980 B packed and 2,009,292 B unpacked.sha512-ompJW0KYgpxj7vDz7W42qCIG6ICj58eAalf17S4cC2AgKTOOJB6A7g8mTBFT+oJOKEjT/WrK+fPZSoTv3ngjWA==.codecarto/engineering/entries and nohosts/node scripts/smoke-mcp.mjs --tarball …gives 12/12 ok (both the 2025initializeera and the2026-07-28server/discoverera).Surface verification (CONTRIBUTING, against the packed tarball's
dist/)Each MCP surface pointed a scratch server name at the installed tarball's
dist/mcp-server/bin.mjs, in a separate copy of a six-file scratch repo.claude -p --strict-mcp-config --mcp-config …:codecarto_init(lite), thencodecarto_nextInitialized CodeCartographer workspace at …/rc/claude/.codecarto.(2)Read .codecarto/GUIDE.md and continue the CodeCartographer workflow for the phase \architecture`.` (the prompt is ≈2,970 chars, not truncated)codex exec --ephemeral -s read-only, withmcp_servers.ccrel.default_tools_approval_mode="approve"and the installed registration disabledmcp: ccrel/codecarto_init (completed),mcp: ccrel/codecarto_next (completed); (1)Initialized CodeCartographer workspace at …/rc/codex/.codecarto.(2)Read .codecarto/GUIDE.md and continue the CodeCartographer workflow for the phase \architecture`.`hermes mcp add ccrel …(the handshake found 23/23 tools), thenhermes chat -Q --oneshot; the server was removed afterwards and the config diff against a backup is identicalInitialized CodeCartographer workspace at …/rc/hermes/.codecarto.(2)Read .codecarto/GUIDE.md and continue the CodeCartographer workflow for the phase \architecture`.`script -qec "pi -ne -e <tarball>/dist/extensions/codecarto/index.js --model openai-codex/gpt-5.5 -p '/codecarto-init architecture-only' '/codecarto-next --auto'"Phase architecture sub-agent finished (36 tool uses, 12 turns).,Auto pipeline complete: 1/1 phases.;current_phase: complete; the artifact's**Overall:** PASS WITH GAPS; the gap was routed toq-cli-distribution-intentThe full round trip ran on Pi: init → next → artifact and handoff written by the model → validate → complete. The real model's handoff passed the new #453 shape gates and its open question reached
status.yamlwith its description. No repair turn was needed, so the #454 repair path was not exercised live; it is covered by unit tests with a fake session only. Nocarry_forwardwas routed, because architecture-only has no downstream phase.Pi model note: the CONTRIBUTING model (
openrouter/google/gemini-3.7-flash) could not be used because the OpenRouter key on this machine is expired (401).openai-codex/gpt-5.3-codex-sparkis refused for ChatGPT-account Codex logins, so the run usedopenai-codex/gpt-5.5.Broad-Side: not touched since 0.26.0, so the live Broad-Side check does not apply.
Known, not blocking
nextaction for the traverse planner.engineering/to.codecarto/.gitignoreby hand; the CHANGELOG says so.