444069d601
* fix(#3613): copy component dirs into the plugin-validate fixture C2 builds its synthetic plugin root by symlinking commands/, hooks/ and skills/ into a temp dir. `claude plugin validate` (>=2.1.233) reads component directories without following symlinks and warns on each one, and `--strict` promotes a warning to a non-zero exit — so the assertion failed on how the fixture was built, not on the manifest under test. Copy the directories with fs.cpSync instead. The validated tree still holds only plugin.json and the three component directories, so nothing else from the repo root is placed where the validator can read it, and the #2665 CLI-config isolation is untouched. The construction helper is self-cleaning: its callers' try/finally only begins once it returns, so a throw partway through would strand a half-built root on disk. The pre-refactor code ran these same steps inside C2's own try, and that teardown guarantee is preserved rather than narrowed. Its cleanup is best-effort so a teardown error cannot replace the real construction error. Add C3, an unconditional tripwire asserting the fixture exposes real, non-symlinked component directories at any depth. C2 never runs in CI — no job under .github/workflows/ installs the `claude` binary — so a revert to symlinks would pass every CI lane and surface only as a red suite on contributor machines. C3 is not a substitute for C2's end-to-end check and cannot be shown red against this base, since the helper it calls arrives with it; C2 is the failing-first artifact. C3 enforces a symlink-free fixture throughout, deliberately stricter than the CLI's own boundary. Measured on 2.1.234, `plugin validate --strict` exits 1 for a symlinked component dir and for a symlink one level inside skills/, and 0 for two levels in or for symlinks under commands/ or hooks/. Encoding that external, undocumented line would be more fragile than a superset costing one walk over ~176 files. The walk uses an explicit stack rather than readdirSync's `recursive: true`, which follows directory symlinks — a link pointing outward would otherwise traverse an unrelated tree, or a cycle, before the assertion ran. Correct the Section C docstring, which claimed C2 provides defence-in-depth coverage it cannot provide in CI. Whether to provision the CLI in a CI job is a maintainer call and is left open. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(#3613): skip hooks/dist and .dist-staging when building the fixture The recursive copy added for #3613 races the hook builders. build-hooks.js writes atomically through a per-PID hooks/.dist-staging-<pid> and removes it when done, and nine test files invoke that script from their before() hooks, so a recursive walk can enumerate a staging directory and then lstat it after the owning process deleted it — the ENOENT that #3656 just fixed in the cold-tree fixture. Copy entry-by-entry and skip by NAME BEFORE anything stats it, reusing shouldCopyHookEntry from tests/helpers/cold-runtime-lib-fixture.cjs rather than re-deriving the rule. A filter applied after the stat would not close it. The predicate is already pinned including its over-match cases, which a loose startsWith('dist') would get wrong: dist-staging-no-dot and distant.js must both be kept. Excluding hooks/dist is independently right for this fixture — a real marketplace install contains neither dist nor a transient staging dir, the same reasoning that keeps the repo-root CLAUDE.md out of the validated tree. C2 still validates clean without it. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(#3613): validate agents/ too, and scope the hooks predicate Three review Minors. agents/ ships in package.json `files` and IS auto-validated by the CLI — verified: a frontmatter-less agents/*.md exits 1. Including it was pointless while the fixture symlinked, because the CLI read nothing through a symlink; now that the tree is real it is the last shipped component tree C2 could not see. Proven to buy coverage rather than bytes: planting a frontmatter-less agent now reds C2, which it could not do before this change. shouldCopyHookEntry is documented as a hooks/ entry filter, so it now runs only for hooks/. Applying it to the other trees was harmless today but silently encoded a hooks-shaped exclusion into them — a future commands/dist would have vanished from the validated tree with no signal. assert.deepEqual -> deepStrictEqual on the nested-symlink assertion. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(#3613): scope the fixture to the three directories the issue names Review findings, all five. Major 1 — `agents/` dropped from COMPONENT_DIRS. The observation behind adding it was right (the CLI does auto-validate it; a frontmatter-less agents/*.md exits 1), but it is a NEW gate the issue does not ask for, on the largest of the trees, and since C2 never runs in CI it would be red only on contributor machines with `claude` installed — the same worst-of-both-states #3613 exists to remove. Worth having as its own issue, where "should CI provision the CLI" gets answered for it too. Major 2 — C3 no longer passes on a fixture that validates nothing. buildValidationPluginRoot() mkdir's every component dir before the entry loop, so a copy that stops happening leaves real, EMPTY directories and all three structural assertions still hold (an empty tree contains no symlinks). Adds a non-empty assertion plus a known entry per tree, so a partial copy is caught as well. Stubbing the copy now reds C3 with "commands/ is EMPTY"; dropping just commands/gsd reds it too. Neither did before. Minor 1 — the measured table was wrong, and the mistake was measuring an inert file. Re-measured on CLI 2.1.239, reproducing the review's 2.1.237 result: a symlinked skills/<name>/SKILL.md at depth 2 exits 1, while a stray symlinked *.md at the same depth exits 0. The boundary is not depth at all, it is whether the symlink is a file the CLI reads as a component. Table replaced with that. Minor 2 — the hooks name filter is now local instead of importing cold-runtime-lib-fixture.cjs's, whose docstring scopes it to the cold-tree fixture and which had exactly one caller. Two consumers across fixtures with different requirements and nothing asserting they stay compatible is how a later cold-tree change silently alters what this fixture validates. Nit 1 — cost figures re-measured: ~1.06 MB over 176 files, not 2.1 MB over 243. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(#3613): correct the hooks row and the count the agents/ removal left behind Three comment-prose items from review, no code change. N1 — the corrected table gained a new wrong row, erring unsafe. "symlink inside commands/ or hooks/ -> 0" is false for hooks/hooks.json, which is the single most likely thing anyone would symlink there. Re-measured on CLI 2.1.239, matching the review's 2.1.237 figures on every cell: symlink inside commands/ (a dir, or a component *.md) 0 stray symlinked dir or *.js inside hooks/ 0 symlinked hooks/hooks.json 1 hooks/hooks.json is now its own row, and is named in the C3 assertion message alongside SKILL.md — that message is what the next engineer actually reads. N2 — "the four component directories" was residue from the revision that also copied agents/; it survived the very edit that removed it, three lines above a sentence saying three. Now three. N3 — the promised agents/ follow-up is filed as #3751 and the comment cites it, rather than promising an issue that did not exist. It frames the real decision (should CI provision the claude CLI) rather than just asking for the directory back. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-authored-by: Tom Boucher <trekkie@nomorestars.com>
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, Antigravity CLI, Kimi CLI, Kilo, Codex, Copilot, Cursor, Windsurf, and more.
What is GSD Core
GSD Core is a context-engineering and spec-driven development framework that drives AI coding agents (Claude Code, Codex, Antigravity 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:
- Discuss — capture implementation decisions before anything is planned
- Plan — research, decompose, and verify the plan fits a fresh context window
- Execute — run plans in parallel waves; each executor starts with a clean 200k-token context
- Verify — walk through what was built; diagnose and fix before declaring done
- 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, Antigravity 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 a new project or onboard an existing repo:
/gsd-new-project # greenfield project
/gsd-onboard # existing codebase
New here? Follow Your first project for a guided walkthrough from install to first shipped phase, or Onboarding an existing codebase for brownfield setup.
Documentation
What's new in 1.7.0 → docs/whats-new-1.7.0.md
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
License
MIT License. See LICENSE for details.
Claude Code is powerful. GSD Core makes it reliable.