Tom Boucher 46967baae8 fix(#948): guard STATE.md no-op writes; preserve milestone_name/stopped_at (closes #944) (#952)
* test(#948): add regression tests for no-op write guard and record-session auto-create (#944)

Red before fix: 11/15 tests fail. Green after: 15/15.
Covers zero-match patch byte-identity, milestone_name preservation,
stopped_at frontmatter-wins, record-session auto-create fallback, and
adversarial fixtures (CRLF, empty body, non-canonical labels).

Also registers bug-948-state-noop-write-guard.test.cjs in the state
bucket of lint-test-file-count.allowlist.json.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix(#948): guard STATE.md no-op writes; preserve milestone_name/stopped_at (closes #944)

Shared root cause: `readModifyWriteStateMd` wrote STATE.md unconditionally
even when the transform produced no change, and `syncStateFrontmatter`
re-derived frontmatter from the possibly-stale body on every write.

Three coordinated fixes in src/state.cts:

1. readModifyWriteStateMd: add no-op guard — when transform result ===
   input content, skip the write entirely (no platformWriteSync, no
   last_updated bump, no frontmatter re-derive). Fixes #948 zero-match
   phantom write and the #944 phantom last_updated bump.

2. syncStateFrontmatter: extend existing-frontmatter preserve logic —
   fall back to existingFm['milestone_name'] / existingFm['milestone']
   when the derived value is the template placeholder 'milestone'
   (getMilestoneInfo returns this literal when it cannot match the
   version in ROADMAP.md); prefer existingFm['stopped_at'] /
   existingFm['paused_at'] over a body-derived value (the frontmatter
   value, written by the canonical record-session path, wins over stale
   historical body lines). Mirrors the fallback already in cmdStateJson.

3. cmdStateRecordSession: when --stopped-at / --resume-file are supplied
   but body labels are absent, DWIM auto-create a canonical ## Session
   section (mirroring how add-decision / add-blocker / record-metric
   auto-create their sections). Never return a silent recorded:false when
   the caller supplied values.

SDK check: no sdk/src/state.ts exists in this repo (the comment in
cmdStateSnapshot references a sibling concern in the TypeScript SDK
codebase, which is a separate repo not present here).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* chore: add changeset for PR #952 (fix #948/#944)

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix(#948): correct stopped_at preserve rule; adjust test for sync behaviour

The "always prefer frontmatter stopped_at" rule in syncStateFrontmatter
was too aggressive — it broke phase.complete which intentionally updates
stopped_at in the body and expects syncStateFrontmatter to pick it up.

The primary fix (no-op guard in readModifyWriteStateMd) already prevents
the stale-body-overwrites-frontmatter scenario from #948: the file is not
written when the transform produces no change, so syncStateFrontmatter
never runs on a zero-match patch. The body-derived value can only win when
an actual write occurs, which means the body was legitimately updated.

Reverted to the original #905 rule for stopped_at/paused_at: fall back to
existing frontmatter only when the derived value is absent (empty/null).

Also adjusted the sync-suite test to assert what state sync actually does
(milestone_name preservation) rather than a stopped_at-wins property that
state sync does not have by design.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix(#944): update existing session block in place (adversarial review)

HIGH finding: the DWIM auto-create in cmdStateRecordSession was appending
a second ## Session block unconditionally, even when one already existed
with non-canonical content (e.g. a markdown table). Both
buildStateFrontmatter and cmdStateSnapshot read only the FIRST ## Session
block via regex, so the newly-written Stopped at / Resume file values
landed in the second, invisible block — frontmatter stopped_at stayed
stale and state-snapshot returned nulls.

Fix: check for an existing ## Session heading. When one is present,
normalize that section in place by replacing its body with canonical
**Last session:** / **Stopped at:** / **Resume file:** bold-label lines.
Only append a brand-new section when NO ## Session heading exists.

LOW finding: the auto-create scaffold emits **Last session:** but
cmdStateSnapshot only matched **Last Date:**, so session.last_date was
null after auto-create despite a valid timestamp being written.

Fix: extend the lastDateMatch regex in cmdStateSnapshot to also accept
**Last session:** / Last session: (the form the scaffold writes).

Tests: 3 new tests added to bug-948-state-noop-write-guard.test.cjs that
confirmed failure against the previous HEAD and pass after this fix:
- exactly one ## Session block after record-session with non-canonical existing block
- state-snapshot sees correct stopped_at via first Session block (not a duplicate)
- state-snapshot session.last_date is non-null after auto-create on body-less file

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix(#944): improve in-place section replace to cleanly remove old body content

The previous regex `/(^## Session[ \t]*$)([\s\S]*?)(?=\n^## |\n*$)/im`
with a lazy match consumed nothing after the heading, so old non-canonical
body content (e.g. table rows) remained after the new canonical lines.

While functionally correct (parsers found the canonical lines first in the
FIRST ## Session block), it left stale content in the section. Replace with
a negative-lookahead per-line pattern that consumes all content from the
heading up to (but not including) the next ## heading, producing a clean
section with only the canonical bold-label lines.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---------

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-10 00:22:29 -04:00

GSD Core

Git. Ship. Done.

English · Português · 简体中文 · 日本語 · 한국어

A light-weight meta-prompting, context engineering, and spec-driven development system for Claude Code, OpenCode, Gemini CLI, Kimi CLI, Kilo, Codex, Copilot, Cursor, Windsurf, and more.

npm version npm downloads Tests Discord GitHub stars License


What is GSD Core

GSD Core is a context-engineering and spec-driven development framework that drives AI coding agents (Claude Code, Codex, Gemini CLI, Kimi CLI, Copilot, Cursor, and more) through a disciplined phase loop. It solves context rot — the quality degradation that accumulates as an AI fills its context window — by running all heavy research, planning, and execution work in fresh-context subagents while keeping your main session lean.


How it works

Each milestone repeats the same five-step loop, one phase at a time:

  1. Discuss — capture implementation decisions before anything is planned
  2. Plan — research, decompose, and verify the plan fits a fresh context window
  3. Execute — run plans in parallel waves; each executor starts with a clean 200k-token context
  4. Verify — walk through what was built; diagnose and fix before declaring done
  5. Ship — create the PR, archive the phase, repeat for the next one

Quickstart

npx @opengsd/gsd-core@latest

The installer prompts for your runtime (Claude Code, OpenCode, Gemini CLI, Kimi CLI, Kilo, Codex, Copilot, Cursor, Windsurf, and more) and whether to install globally or locally. The installer is required for cross-runtime compatibility — do not copy files from agents/ or commands/ directly.

On another runtime or without Node.js? See Install on your runtime.

Once installed, start your first project:

/gsd-new-project

New here? Follow Your first project for a guided walkthrough from install to first shipped phase.


Documentation

Tutorials — learning by doing:

How-to guides — task-focused recipes:

Reference — authoritative facts:

Explanation — concepts and design decisions:

Full index: docs/README.md. Other languages: 日本語 · 한국어 · Português · 简体中文.


Why it works

Most AI-coding setups fail at scale because context bloat silently degrades output quality, there is no shared memory between sessions, and nothing verifies that code actually works. GSD Core solves all three: heavy work runs in fresh subagents, structured artifacts like STATE.md and CONTEXT.md survive session boundaries, and the verify step walks through what was built and generates fix plans before a phase is declared done. See docs/explanation/context-engineering.md for the full reasoning.

Troubleshooting? See docs/how-to/recover-and-troubleshoot.md.


Community

Project Platform
gsd-opencode Original OpenCode port
Discord Community support

Star History

Star History Chart

License

MIT License. See LICENSE for details.


Claude Code is powerful. GSD Core makes it reliable.

Description
No description provided
Readme MIT 77 MiB
Languages
JavaScript 82.3%
TypeScript 17.4%
Shell 0.3%