Read context_window config (default 200000) in execute-phase and plan-phase workflows. When >= 500000 (1M-class models), subagent prompts include richer context: executor agents receive CONTEXT.md, RESEARCH.md, and prior wave SUMMARYs; verifier agents receive all PLANs, SUMMARYs, and REQUIREMENTS.md; planner receives prior phase CONTEXT.md for cross-phase decision consistency.
At 200k (default), behavior is unchanged.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Both the gsd-executor and gsd-verifier Task() invocations were missing
the required `description` parameter, causing InputValidationError when
spawning agents in parallel during /gsd:execute-phase.
Reject project/workstream names containing path separators or ..
components. Covers both GSD_PROJECT and GSD_WORKSTREAM. Adds 9 tests
for the full resolution matrix and traversal rejection cases.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds OpenCode CLI as a 5th reviewer option, enabling teams with GitHub
Copilot subscriptions to leverage Copilot-routed models (e.g.
gpt-5.3-codex) for cross-AI plan reviews.
Closes#1520
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add check-commit command to gsd-tools that acts as a pre-commit guard.
When commit_docs is false, rejects commits that stage .planning/ files
with an actionable error message including the unstage command.
Recreated cleanly on current main — previous version carried stale
shared fixes that are now upstream.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add --diagnose flag to /gsd:debug that stops after finding the root
cause without applying a fix. Returns a structured Root Cause Report
with confidence level, files involved, and suggested fix strategies.
Offers "Fix now" to spawn a continuation agent, "Plan fix", or
"Manual fix" options.
Recreated cleanly on current main — previous version carried stale
shared fixes that are now upstream.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The autonomous workflow previously had two extremes: full auto-answer
mode (--auto discuss, lean context but no user input) or manager mode
(interactive but bloated context from accumulating everything inline).
The --interactive flag bridges this gap:
- Discuss runs inline via gsd:discuss-phase (asks questions, waits for
user answers — preserving all design decisions)
- Plan and execute dispatch as background agents (fresh context per
phase — no accumulation in the main session)
- Pipeline parallelism: discuss Phase N+1 while Phase N builds in the
background
This keeps the main context lean (only discuss conversations accumulate)
while preserving user input on all decisions. Particularly helpful for
users hitting context limits with /gsd:manager on multi-phase milestones.
Usage:
/gsd:autonomous --interactive
/gsd:autonomous --interactive --from 3
/gsd:autonomous --interactive --only 5
Also adds --only N flag parsing to the upstream workflow (previously only
in PR #1444's branch).
Closes#1413
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add tests covering the three manual verification scenarios for the
context_window and workflow.subagent_timeout config features:
- init execute-phase output includes context_window from config (both
custom 1M value and 200k default)
- config-get context_window returns the configured value (and errors
when absent)
- config-set workflow.subagent_timeout accepts numeric values with
proper string-to-number coercion and round-trips through config-get
All 1517 tests pass.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The map-codebase workflow had a hardcoded 300000ms (5 minute) timeout
for parallel subagent tasks. On large codebases or with slower models
(e.g. GPT via Codex), subagents can need 10-20+ minutes, causing the
parent to kill still-working agents and fall back to sequential mode.
Changes:
- Add workflow.subagent_timeout config key (default: 300000ms)
- Register in VALID_CONFIG_KEYS (config.cjs)
- Add to loadConfig() defaults and return object (core.cjs)
- Emit in map-codebase init context (init.cjs)
- Update map-codebase.md to use config value instead of hardcoded 300000
- Document in planning-config.md reference
Users can now increase the timeout via:
/gsd:settings workflow.subagent_timeout 900000
Closes#1472
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add two config-get tests to verify the git.base_branch round-trip:
- config-get returns the value after config-set stores it
- config-get errors with "Key not found" when git.base_branch is not
explicitly set (default config omits it), which triggers the
auto-detect fallback via origin/HEAD in workflows
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds `git.base_branch` config option that controls the target branch
for PRs and merges. When unset, auto-detects from origin/HEAD and
falls back to "main".
This fixes projects using `master`, `develop`, or any other default
branch — previously `/gsd:ship` would create PRs targeting `main`
(which may not exist) and `/gsd:complete-milestone` would try to
checkout `main` and fail.
Changes:
- config.cjs: add git.base_branch to valid config keys
- planning-config.md: document the option with auto-detect behavior
- ship.md: detect base branch at init, use in PR create, branch
detection, push report, and completion report
- complete-milestone.md: detect base branch, use for squash merge
and merge-with-history checkout targets
- 1 new test for config-set git.base_branch
Usage:
gsd-tools config-set git.base_branch master
Or auto-detect (default — reads origin/HEAD):
git.base_branch: null
Closes#1466
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
permissionMode: acceptEdits in gsd-executor and gsd-debugger frontmatter
is Claude Code-specific and causes Gemini CLI to hard-fail on agent load
with "Unrecognized key(s) in object: 'permissionMode'". The field also
has no effect in Claude Code (subagent Write permissions are controlled
at runtime level regardless). Remove it from both agents and update
tests to enforce cross-runtime compatibility.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When phase numbers are globally sequential across milestones (e.g.,
phases 61-67), the banner showed "Phase 63/5" where 5 was the count of
remaining phases — easily mistaken for "63 out of 5 total." Clarify
that T must be total milestone phases and add fallback display format
for when phase numbers exceed the total count.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
workstream set with no argument silently cleared the active workstream,
a footgun for users who forgot the name. Now requires a name arg and
errors with usage hint. Explicit clearing via --clear flag, which also
reports the previous workstream in its output.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add --augment flag to install for Augment only
- Add Augment to --all and interactive menu (now 9 runtimes + All)
- Create conversion functions for Augment skills and agents:
- convertClaudeToAugmentMarkdown: path and brand replacement
- convertClaudeCommandToAugmentSkill: skill format with adapter header
- convertClaudeAgentToAugmentAgent: agent format conversion
- copyCommandsAsAugmentSkills: copy commands as skills
- Map tool names: Bash → launch-process, Edit → str-replace-editor, etc.
- Add runtime label and uninstall support for Augment
- Add tests: augment-conversion.test.cjs with 15 test cases
- Update multi-runtime-select.test.cjs to include Augment
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Claude Code 2.1.88+ deprecated commands/ subdirectory discovery in favor of
skills/*/SKILL.md format. This migrates the Claude Code installer to use the
same skills pattern already used by Codex, Copilot, Cursor, Windsurf, and
Antigravity.
Key changes:
- New convertClaudeCommandToClaudeSkill() preserving allowed-tools and argument-hint
- New copyCommandsAsClaudeSkills() mirroring Copilot pattern
- Install now writes skills/gsd-*/SKILL.md instead of commands/gsd/*.md
- Legacy commands/gsd/ cleaned up during install
- Manifest tracks skills/ for Claude Code
- Uninstall handles both skills/ and legacy commands/
Fixes#1504
Supersedes #1538
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds --chain flag to /gsd:discuss-phase that provides the middle ground
between fully manual and fully automatic workflows:
/gsd:discuss-phase 5 --chain
- Discussion is fully interactive (user answers questions)
- After context is captured, auto-advances to plan → execute
- Same pipeline as --auto, but without auto-answering
This addresses the community request for per-phase automation where
users want to control discuss decisions but skip manual advancement
between plan and execute steps.
Workflow: discuss (interactive) → plan (auto) → execute (auto)
Changes:
- Workflow: --chain flag triggers auto_advance without auto-answering
- Workflow: chain flag synced alongside --auto in ephemeral config
- Workflow: next-phase suggestion preserves --chain vs --auto
- Command: argument-hint and description updated
- Success criteria updated
Closes#1327
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
