Drill scenario worktree-already-inside-spec-aware revealed a latent
skill loophole: agents detecting existing isolation would correctly
refuse to nest a worktree, but then offer options — including exiting
to the main repo and creating a sibling worktree from there — and
execute the sibling option when the actor accepted.
The prior wording "Do NOT create another worktree" got read as
"don't NEST another worktree," leaving sibling creation as an
apparently-legal workaround that satisfied the agent's interpretation
of the directive while violating the skill's actual intent (use the
current workspace, period).
Fix: tighten the already-inside block to be explicit:
- Current workspace is the working environment, use it
- No alternatives, no creation — not nested, not sibling, not anywhere
- If the user genuinely needs different isolation for unrelated work,
they have to exit the current workspace themselves
Validation pending via drill reruns of worktree-already-inside and
worktree-already-inside-spec-aware on this branch state.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The skill previously required an explicit reply to its "do you want a
worktree?" dialogue, which produced obtuse UX when the user invoked the
skill by name — agents had to stop and ask "do you want a worktree?"
even though the user just asked for the skill whose purpose is worktrees.
Loosen Step 2 to recognize the invoking turn as consent: if the user's
most recent message named the skill, asked for a worktree, or asked for
an isolated workspace, proceed directly to Step 3 without re-prompting.
The gate still fires for the transitive case (agent infers isolation
from a feature description) — that remains the #991 failure mode.
Also trim "or skill invocation" from the anti-inference Red Flag and
destale the Integration section now that SDD/executing-plans no longer
require a worktree.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Post-inversion cleanup:
- executing-plans, subagent-driven-development: update Integration
description from "Ensures isolated workspace" to "Detects workspace
environment and offers worktree isolation on request"
- codex-tools.md: update step references (Step 0→1, Step 1→2)
- using-git-worktrees Step 2: restore "silence → ask once more" instead
of "silence → work in place" to preserve safety net for confused users
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Agents consistently skipped Step 0.5 (consent gate) because fractional
numbering signals "optional afterthought" and the prose-only step was
invisible to code-block anchoring. The fix inverts the structural
gravity: the default path now works in place, and worktree creation
is an off-ramp requiring explicit user request.
- Renumber to clean integers: Step 1 (detect) → 2 (offer) → 3 (create) → 4 (setup) → 5 (verify)
- Step 2 defaults to Step 4 (in-place); Step 3 only on explicit user ask
- Step 2 includes a code block so agents register it during execution
- Add "creating without being asked" to Common Mistakes
- Add anti-inference red flag: consent from task/plan/skill invocation doesn't count
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Drill benchmark showed 0/4 consent compliance across both Claude Code
and Codex. Root cause: consent was buried inline in Step 0's conditional
branch. Agents anchor on the next bash command and skip prose.
Fix: promote consent to its own numbered section with imperative framing
("REQUIRED STOP", "Do NOT proceed without an answer") and exact output
template. Also adds explicit "no" path — users who want to work directly
on their current branch skip to Step 3 with no worktree creation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Addresses review feedback on #1123. Replaces "(same Step 0→1a→1b flow,
matching branch names)" with plain-language instruction that doesn't
forward-reference section numbers that could rot under future edits.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Jesse flagged that the verbose CLAUDE.md/AGENTS.md/GEMINI.md/.cursorrules
enumeration (a) chews tokens, (b) confuses models that anchor on exact
strings, and (c) is repeated DRY-violatingly across 3+ locations.
Replace with abstract "your instructions" framing in four spots:
- skills/using-git-worktrees/SKILL.md Step 0 → Step 1 transition
- skills/using-git-worktrees/SKILL.md Step 1b Directory Selection
- docs/superpowers/plans/2026-04-06-worktree-rototill.md (both mirror locations)
Same intent, harness-agnostic phrasing, ~half the tokens.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Address obra's comment on explicit step numbers / prescriptive tone.
Drops "STOP HERE if available", the "If YES:" gate, and the "even if /
even if / NO EXCEPTIONS" reinforcement paragraph. Keeps the specific
tool-name anchors (EnterWorktree, WorktreeCreate, /worktree, --worktree),
which the original TDD data showed are load-bearing.
A/B verified against drill harness on the 3 creation/consent scenarios
(consent-flow, creation-from-main, creation-from-main-spec-aware):
baseline explicit wording scored 12/12 criteria, softened wording also
scored 12/12. The "agent used the most appropriate tool" criterion
passed in all 3 softened runs — agents still picked EnterWorktree via
ToolSearch without the imperative framing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Consent prompt: drop "(y/n)" and add escape valve for users who
have already declared their worktree preference in global or
project agent instruction files.
- Directory selection: reorder to put declared user preference
ahead of observed filesystem state, and reframe the default as
"if no other guidance available".
- Sandbox fallback: require explicitly informing the user that
the sandbox blocked creation, not just "report accordingly".
- writing-plans: fully qualify the superpowers:using-git-worktrees
reference.
- Plan doc: mirror the consent-prompt change.
Step 1a native-tool framing and the helper-scripts suggestion are
still outstanding — the first needs a benchmark re-run before softer
phrasing can be adopted without regressing compliance; the second is
exploratory and will get a thread reply.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Git worktrees inherit hooks from the main repo automatically via
$GIT_COMMON_DIR — this has been the case since git 2.5 (2015).
The symlink step was based on an incorrect premise from PR #965
and also fails in practice (.git is a file in worktrees, not a dir).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Step 1a failed at 2/6 with the spec's original abstract text ("use your
native tool"). Three REFACTOR iterations found what works (50/50 runs):
1. Explicit tool naming — "do you have EnterWorktree, WorktreeCreate..."
transforms interpretation into factual toolkit check
2. Consent bridge — "user's consent is your authorization" directly
addresses EnterWorktree's "ONLY when user explicitly asks" guardrail
3. Red Flag entry naming the specific anti-pattern
File split was tested but proven unnecessary — the fix is the Step 1a
text quality, not physical separation of git commands. Control test
with full 240-line skill (all git commands visible) passed 20/20.
Test script updated: supports batch runs (./test.sh green 20), "all"
phase, and checks absence of git worktree add (reliable signal) rather
than presence of EnterWorktree text (agent sometimes omits tool name).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The creation skill supports both .worktrees/ and worktrees/ directories,
but the finishing skill's cleanup only checked .worktrees/. Worktrees
under the non-hidden path would be orphaned on merge or discard.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Remove REQUIRED language from executing-plans and subagent-driven-development.
Consent and detection now live inside using-git-worktrees itself.
Fix stale 'created by brainstorming' claim in writing-plans.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
using-git-worktrees: submodule guard now says "treat as normal repo"
instead of "proceed to Step 1" (preserves consent flow)
using-git-worktrees: directory priority summaries include global legacy
finishing-a-development-branch: move git branch -d after Step 6 cleanup
to make Bug #999 ordering unambiguous (merge -> worktree remove -> branch delete)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Step 2: environment detection (GIT_DIR != GIT_COMMON) before presenting menu
Detached HEAD: reduced 3-option menu (no merge from detached HEAD)
Provenance-based cleanup: .worktrees/ = ours, anything else = hands off
Bug #940: Option 2 no longer cleans up worktree
Bug #999: merge -> verify -> remove worktree -> delete branch
Bug #238: cd to main repo root before git worktree remove
Stale worktree pruning after removal (git worktree prune)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Two bugs caused the brainstorm server to self-terminate within 60s:
1. ownerAlive() treated EPERM (permission denied) as "process dead".
When the owner PID belongs to a different user (Tailscale SSH,
system daemons), process.kill(pid, 0) throws EPERM — but the
process IS alive. Fixed: return e.code === 'EPERM'.
2. On WSL, the grandparent PID resolves to a short-lived subprocess
that exits before the first 60s lifecycle check. The PID is
genuinely dead (ESRCH), so the EPERM fix alone doesn't help.
Fixed: validate the owner PID at server startup — if it's already
dead, it was a bad resolution, so disable monitoring and rely on
the 30-minute idle timeout.
This also removes the Windows/MSYS2-specific OWNER_PID="" carve-out
from start-server.sh, since the server now handles invalid PIDs
generically at startup regardless of platform.
Tested on Linux (magic-kingdom) via Tailscale SSH:
- Root-owned owner PID (EPERM): server survives ✓
- Dead owner PID at startup (WSL sim): monitoring disabled, survives ✓
- Valid owner that dies: server shuts down within 60s ✓
Fixes#879
ownerAlive() treated EPERM (permission denied) the same as ESRCH
(process not found), causing the server to self-terminate within 60s
whenever the owner process ran as a different user. This affected WSL
(owner is a Windows process), Tailscale SSH, and any cross-user
scenario.
The fix: `return e.code === 'EPERM'` — if we get permission denied,
the process is alive; we just can't signal it.
Tested on Linux via Tailscale SSH with a root-owned grandparent PID:
- Server survives past the 60s lifecycle check (EPERM = alive)
- Server still shuts down when owner genuinely dies (ESRCH = dead)
Fixes#879
The session directory now contains two peers: content/ (HTML served to
the browser) and state/ (events, server-info, pid, log). Previously
all files shared a single directory, making server state and user
interaction data accessible over the /files/ HTTP route.
Also fixes stale test assertion ("Waiting for Claude" → "Waiting for
the agent").
Reported-By: 吉田仁
Metadata files (.server-info, .events, .server.pid, .server.log,
.server-stopped) were stored in the same directory served over HTTP,
making them accessible via the /files/ route. They now live in a .meta/
subdirectory that is not web-accessible.
Also fixes a stale test assertion ("Waiting for Claude" → "Waiting for
the agent").
Reported-By: 吉田仁
The subagent review loop (dispatching a fresh agent to review plans/specs)
doubled execution time (~25 min overhead) without measurably improving plan
quality. Regression testing across 5 versions (v3.6.0 through v5.0.4) with
5 trials each showed identical plan sizes, task counts, and quality scores
regardless of whether the review loop ran.
Changes:
- writing-plans: Replace subagent Plan Review Loop with inline Self-Review
checklist (spec coverage, placeholder scan, type consistency)
- writing-plans: Add explicit "No Placeholders" section listing plan failures
(TBD, vague descriptions, undefined references, "similar to Task N")
- brainstorming: Replace subagent Spec Review Loop with inline Spec Self-Review
(placeholder scan, internal consistency, scope check, ambiguity check)
- Both skills now use "look at it with fresh eyes" framing
Testing: 5 trials with the new skill show self-review catches 3-5 real bugs
per run (spawn positions, API mismatches, seed bugs, grid indexing) in ~30s
instead of ~25 min. Remaining defects are comparable to the subagent approach.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Skip OWNER_PID monitoring on Windows/MSYS2 where the PID namespace is
invisible to Node.js, preventing server self-termination after 60s (#770)
- Document run_in_background: true for Claude Code on Windows (#767)
- Restore user choice between subagent-driven and inline execution after
plan writing; subagent-driven is recommended but no longer mandatory
- Add Windows lifecycle test script verified on Windows 11 VM
- Note #723 (stop-server.sh reliability) as already fixed
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Remove chunk-based plan review in favor of single whole-plan review
- Add Calibration sections to both reviewer prompts so only serious
issues block approval
- Reduce max review iterations from 5 to 3
- Streamline reviewer checklists (spec: 7→5, plan: 7→4 categories)
Replace #!/bin/bash with #!/usr/bin/env bash in 13 scripts. The
hardcoded path fails on NixOS, FreeBSD, and macOS with Homebrew bash.
#!/usr/bin/env bash is the portable POSIX-friendly alternative.
Tested on Linux and Windows (Git Bash + CMD). macOS is the primary
beneficiary since Homebrew installs bash to /opt/homebrew/bin/bash.
Based on #700, closes#700.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Windows/Git Bash reaps nohup background processes, causing the brainstorm
server to die silently after launch. Auto-detect Windows via OSTYPE
(msys/cygwin/mingw) and MSYSTEM env vars, switching to foreground mode
automatically. Tested on Windows 11 from CMD, PowerShell, and Git Bash —
all route through Git Bash and hit the same issue.
Based on #740, fixes#737. Also adds CHANGELOG.md documenting the fix and
a known OWNER_PID/WINPID mismatch on the main branch.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Subagents should never inherit the parent session's context or history.
The dispatcher constructs exactly what each subagent needs, keeping
both sides focused: the subagent on its task, the controller on
coordination.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
$PPID inside start-server.sh is the ephemeral shell the harness spawns
to run the script — it dies immediately when the script exits, causing
the server to shut down after ~60s. Now resolves grandparent PID via
`ps -o ppid= -p $PPID` to get the actual harness process (e.g. claude).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
start-server.sh passes $PPID as BRAINSTORM_OWNER_PID to the server.
The server checks every 60s if the owner process is still alive
(kill -0). If it's gone, the server shuts down immediately —
deletes .server-info, writes .server-stopped, exits cleanly.
Works across all harnesses (CC, Codex, Gemini CLI) since it
tracks the shell process that launched the script, which dies
when the harness dies.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Server tracks activity (HTTP requests, WebSocket messages, file
changes) and exits after 30 minutes of inactivity. On exit, deletes
.server-info and writes .server-stopped with reason. Visual companion
guide now instructs agents to check .server-info before each screen
push and restart if needed. Works on all harnesses, not just CC.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Delete 717 files: index.js, package.json, package-lock.json, and
the entire node_modules directory (express, ws, chokidar + deps).
Update start-server.sh to use server.js. Remove gitignore exception.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Complete zero-dep brainstorm server. Uses knownFiles set to
distinguish new screens from updates (macOS fs.watch reports
'rename' for both). All 56 tests pass (31 unit + 25 integration).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Implements RFC 6455 handshake, frame encoding/decoding for text
frames. All 31 unit tests pass.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The spec review loop (dispatch spec-document-reviewer subagent, iterate
until approved) existed in the prose "After the Design" section but was
missing from both the checklist and the process flow diagram. Since agents
follow the diagram and checklist more reliably than prose, the spec review
step was being skipped entirely.
Added step 7 (spec review loop) to the checklist and a corresponding
"Spec review loop" → "Spec review passed?" node pair to the dot graph.
Tested with claude --plugin-dir and claude-session-driver: worker now
correctly dispatches the spec-document-reviewer subagent after writing
the design doc and before presenting to the user for review.
Fixes#677.
start-server.sh now runs npm install if node_modules is missing.
Fixes broken server when superpowers is installed as a plugin (node_modules
are in .gitignore and not included in the clone).
