v1.58.1.0 feat: hermetic local E2E + Conductor prose AskUserQuestion (#2004)

* feat: add shared call-time isConductor() helper

Single source of truth for Conductor host detection in TS consumers
(CONDUCTOR_WORKSPACE_PATH / CONDUCTOR_PORT). Reads the passed env at
call time, not a module-load snapshot, so unit tests can pin the env
inline without Bun --preload (esm-hoist-breaks-env-pin-bootstrap).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* test: harden question-preference-hook harness against ambient Conductor env

runHook copied all of process.env into the hook subprocess, so running the
suite inside Conductor (CONDUCTOR_WORKSPACE_PATH/PORT set) would leak those
markers. Strip them so the existing cases deterministically characterize
NON-Conductor behavior before the Conductor branch lands. Baseline: 15 pass.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat: PreToolUse hook denies AskUserQuestion in Conductor, redirects to prose

Conductor disables native AskUserQuestion and routes through a flaky MCP
variant that returns '[Tool result missing due to internal error]'. The
hook now denies any AUQ call in a Conductor session and instructs the model
to render a prose decision brief instead (transport avoidance, not preference
enforcement) — firing for one-way doors too, with a typed-confirmation
requirement for destructive paths.

Precedence: never-ask auto-decide still wins (user already settled those);
Conductor prose is the fallback for everything else; non-Conductor behavior
is byte-for-byte unchanged. Restructured the per-question loop to compute
eligibility without early-returning so the Conductor branch can run as the
fallback while preserving memoryContext on every exit.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat: Conductor renders AskUserQuestion decisions as prose by default

In Conductor, native AskUserQuestion is disabled and the MCP variant is
flaky, so skills now render every decision as a plain-text prose brief the
user answers by typing a letter — proactively, not as a failure reaction.

- Preamble emits CONDUCTOR_SESSION, gated on != headless so eval/CI inside
  Conductor still BLOCKs instead of rendering prose to nobody.
- AskUserQuestion Format gains a Conductor-default-prose rule (auto-decide
  preferences still apply first; prose decisions log via gstack-question-log
  since PostToolUse never fires), a one-way/destructive typed-confirmation
  rule, and a typed-reply continuation protocol for split chains.
- Regenerated all SKILL.md + ship golden fixtures; bumped affected carve
  skeleton caps to absorb the always-loaded additions.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat: deploy the Conductor AskUserQuestion hook (setup + upgrade migration)

The PreToolUse hook only delivers its Conductor-prose guarantee if it's
installed, but setup skips hook registration in non-interactive (conductor/CI)
setups. Two fixes so layer 3 actually deploys:

- setup: treat a Conductor workspace as an implicit opt-in for the PreToolUse
  hook on the silent fall-through (never overriding an explicit opt-out).
- migration v1.58.0.0: re-register the hook for existing Conductor installs on
  /gstack-upgrade, idempotent and respecting plan_tune_hooks=no.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* test: E2E for Conductor prose + fix auto-decide-preserved GSTACK_HOME bug

- New skill-e2e-conductor-prose (periodic): Conductor env + plan-eng-review
  surfaces a prose decision brief, not a silent skip. Header documents this is
  end-to-end behavior coverage; the deterministic Conductor guard is the
  question-preference-hook unit test (the PTY harness can't register the MCP
  variant — Codex #10).
- Fix the pre-existing bug in auto-decide-preserved: it seeded the never-ask
  preference under GSTACK_HOME=tmpHome but never passed GSTACK_HOME into the
  PTY run, so the spawned claude read the real ~/.gstack and the preference
  was inert (Codex #9). Now passes GSTACK_HOME + CONDUCTOR_WORKSPACE_PATH to
  prove auto-decide still wins over the Conductor prose redirect.
- Register both in touchfiles (periodic tier).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* v1.58.0.0 feat: Conductor renders AskUserQuestion decisions as prose

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* test: strip ambient Conductor env in memory-cache-injection hook harness

Same dev-in-Conductor leak fixed for question-preference-hook: this suite's
runHook copies process.env, so running it inside Conductor flipped the
defer-path memoryContext assertions into the [conductor] prose deny. Strip
CONDUCTOR_* so the cases characterize non-Conductor behavior. (CI is headless,
so this only bit local Conductor runs.)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat: gstack-detach — run agent eval/bench jobs in their own session

Long agent-run jobs (30-60 min evals, benchmarks) die when the harness sends
SIGTERM to a background task's process group on turn boundaries / monitor
stops / interruptions (observed: 'script test:gate terminated by signal
SIGTERM'). gstack-detach runs the command in a fresh session (python3
os.setsid, or setsid on Linux, nohup fallback) so a group SIGTERM can't reach
it, and wraps it in caffeinate -i on macOS so idle-sleep can't kill it either.
Returns immediately; caller polls the logfile. Secrets stay in env, never argv.

The guard test pins the contract: the command runs in a different process
group than the caller and outlives the launching shell.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat: eval:bg* scripts — detached eval runs for agents

Agent-facing convenience scripts that launch the eval suites through
gstack-detach so a harness SIGTERM can't kill a long run. eval:bg (diff-based),
eval:bg:all, eval:bg:gate, eval:bg:periodic — each returns immediately and
streams to /tmp/gstack-evals.log for polling. The plain test:evals / test:e2e
scripts stay foreground for humans.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs: CLAUDE.md — agents must run long evals via gstack-detach

Codifies the detached-execution default: agent-launched eval/benchmark runs go
through bin/gstack-detach (or the eval:bg* scripts) so a harness SIGTERM or
macOS idle-sleep can't kill a 30-60 min run, then poll the log with a
death-aware watcher. Humans keep foreground scripts.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat: harden gstack-detach against all four eval-infra killers

The basic bash detach fixed SIGTERM but a real run on a shared dev box hit
three more killers: cross-worktree API saturation (15-way concurrency x a
sibling worktree mass-timed-out the suite), a silent hang (periodic bun died
with no exit marker), and shared-/tmp log contamination (a concurrent
worktree's agent output bled into the log). Rewrite as a portable python3 tool
that bakes in all four fixes:

- fork + setsid: SIGTERM-proof (own session, survives harness polite-quit)
- caffeinate -i on macOS: no idle-sleep death
- --lock NAME (fcntl, machine-wide): concurrent worktrees SERIALIZE instead of
  saturating the shared model API
- run-scoped default log (~/.gstack-dev/eval-runs/<label>-<slug>-<branch>-<ts>-<pid>):
  no cross-worktree collision/contamination
- --timeout watchdog + a guaranteed '### gstack-detach EXIT=<code> ###' sentinel
  on every terminal path: no silent hang, finished-vs-died always detectable

Guard test pins all four: detached pgid differs + outlives launcher, run-scoped
log path, watchdog EXIT=timeout, and lock serialization (second run WAITS).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat: eval:bg* use run-scoped logs + machine lock + watchdog

Drop the shared /tmp/gstack-evals.log path (the cross-worktree collision that
contaminated a live run) for gstack-detach's run-scoped default, and add the
machine-wide gstack-evals lock (concurrent worktrees serialize, no API
saturation) plus per-tier watchdog timeouts (60/90/120 min). Each eval:bg*
prints its run-scoped log path to poll.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs: wire detached-eval guidance into /ship + correct CLAUDE.md flags

- /ship eval step (sections/tests.md): long eval suites launch via gstack-detach
  (own session, machine lock, EXIT sentinel) so a turn boundary can't kill a
  30+ min run mid-ship — the exact failure observed during this branch's ship.
- CLAUDE.md: correct the now-stale /tmp reference; document the --lock (serialize
  worktrees, no API saturation), --timeout watchdog, run-scoped log, and the
  guaranteed EXIT sentinel the poller breaks on.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* refactor: extract pure promotedEnv() from conductor-env-shim

Single source of truth for GSTACK_* key promotion semantics. The ambient
promoteConductorEnv() becomes a wrapper; behavior-preserving. Needed by the
hermetic env builder which must not mutate process.env.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat: hermetic child-env builder for E2E runners

Allowlist scrub (basics/network/named-auth kept; CONDUCTOR_*, CLAUDE_*,
GSTACK_*, MCP_*, GBRAIN_*, operator credentials dropped), per-runner
extraAllow, overrides merge last, EVALS_HERMETIC=0 byte-identical escape
hatch read at call time (ESM-hoist safe). Sync memoized singleton temp dirs
(<runRoot>/.claude keeps the extractPlanFilePath contract), seeded
.claude.json for non-interactive first run, pid-aware GC of crashed runs.
19 free unit tests.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat: session-runner spawns hermetic children + isolation canaries

claude -p children now get the allowlist-scrubbed env and a gated
--strict-mcp-config (EVALS_HERMETIC=0 restores operator env AND args).
Two gate-tier canaries make the clean room falsifiable: hermetic-canary
asserts env redirect + scrub + zero MCP servers + nonzero API-key cost
from the Bash tool_result (never model prose); hermetic-sentinel plants a
poisoned operator config (user CLAUDE.md + MCP server) and proves the
child cannot see it. Empirically verified on claude 2.1.175: print mode
needs no seed config (the seed serves the PTY path); the child CLI sets
CLAUDECODE for its own tools, so that scrub is pinned in unit tests, not
E2E. hermetic-env.ts joins GLOBAL_TOUCHFILES.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat: PTY runner spawns hermetic claude sessions

launchClaudePty children get the allowlist-scrubbed env, a gated
--strict-mcp-config, and the session exposes hermeticConfigDir for
forensics (hermetic plan files live under <dir>/plans/ and still match
extractPlanFilePath via the /.claude dir-name contract). Seeded trust
state covers repo-cwd sessions; the 15s trust-watcher stays as fallback.
Verified foreground via the plan-mode-no-op gate test.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat: codex/gemini runners spawn hermetic children

Same allowlist scrub as the claude runners, with each provider's auth
surface re-admitted via extraAllow (codex: OPENAI_API_KEY/CODEX_* plus
its tempHome .codex copy; gemini: GEMINI_*/GOOGLE_* with real HOME for
~/.gemini auth). The gemini spawn previously inherited the full operator
env with no env property at all.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat: agent-sdk-runner spawns hermetic children via complete Options.env

The historical 'env: breaks SDK auth' failure was partial-env replacement:
Options.env replaces the child's entire environment, so objects lacking
ANTHROPIC_API_KEY killed auth. Passing the complete hermetic env (key +
PATH + redirected CLAUDE_CONFIG_DIR/GSTACK_HOME) works — validated live
via query() with a Bash tool call (success, real cost, Conductor vars
scrubbed). Per-test opts.env merges last; ambient key mutation still
works because the builder reads process.env at call time.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* test: static tripwire pins hermetic wiring in all five runners

Free-tier invariants: every runner builds child env via hermeticChildEnv,
no raw ...process.env spread at any spawn site, --strict-mcp-config gated
on isHermeticEnabled in both claude runners, and no test callsite passes
the operator env into a runner's override parameter (scoped to runner
calls — unit tests spawning gstack bin scripts directly are exempt).
Mirrors the terminal-agent-pid-identity / server-embedder-terminal-port
tripwire idiom.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* test: refresh codex/factory ship goldens with detached-eval block

a38089aa added the gstack-detach guidance to the ship template and
updated the claude golden; the codex and factory goldens missed the same
16-line block. Regenerated via bun run gen:skill-docs.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs: hermetic local E2E is the default; retire stale SDK env warning

CLAUDE.md now documents the hermetic clean room (allowlist scrub, fresh
seeded CLAUDE_CONFIG_DIR, temp GSTACK_HOME, --strict-mcp-config),
EVALS_HERMETIC=0 as the debug escape hatch, and replaces the 'never pass
env: to runAgentSdkTest' rule with the verified mechanism (partial-env
replacement was the failure; complete env is safe).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix: operational-learning fixture copies lib/jsonl-store.ts with the bin

gstack-learnings-log imports $SCRIPT_DIR/../lib/jsonl-store.ts (hasInjection,
v1.57.5.0) — copying only the bin scripts into the temp fixture broke the
script with exit 1 since then. Latent because diff-based selection rarely
runs this test; surfaced when hermetic-env.ts joined GLOBAL_TOUCHFILES and
selected everything. Reproduced outside the hermetic env to confirm blame.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix: ios-qa daemon scenarios use unique pidfiles under --concurrent

All scenarios shared join(workDir, 'daemon.pid') through a module-scope
workDir binding that beforeEach reassigns mid-flight under bun --concurrent.
First daemon claims; siblings get already_running against the test process's
own always-alive pid and fail in milliseconds — the failure mode seen at
15-way gate concurrency. Per-claim unique pidfiles keep the single-instance
semantics under test.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix: workflow judge re-appends body-carved sections after the marker slice

runWorkflowJudge appended sections/*.md before slicing startMarker..endMarker.
That handles skills that moved their MARKERS into sections (plan-eng,
plan-design) but not document-release, which keeps its markers in the
skeleton and carved the workflow BODY (Steps 2-9 -> sections/release-body.md)
AFTER the endMarker — so the slice dropped it and the judge scored
completeness 2 ('Steps 2-9 are in an external file'). Now any carved section
the marker window excluded is re-appended, so the judge sees the full
workflow the agent executes. document-release: completeness 2->5, clarity
3->4. ship/plan-ceo/plan-eng/plan-design judges unchanged (their section
content is already inside the slice, so the head-dedup skips re-append).

Pre-existing since the v1.57.0.0 carve (#1907); surfaced now because
hermetic-env.ts is a global touchfile that selects every llm-judge test.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* harden: hermetic temp-dir GC grace window + half-seed cleanup

Codex adversarial review (ship) flagged two temp-dir lifecycle edges:
- GC deleted any dead-pid dir; PID reuse could delete a freshly-created dir
  whose original pid exited and was recycled to a live process. Now requires
  BOTH a dead pid AND mtime older than a 1h floor.
- A seed-write failure after mkdir left an unseeded dir named with our live
  pid that this process's GC skips, leaking until exit. Now the partial dir
  is torn down before the (still loud) rethrow.

Two findings left as-is by design: HOME stays allowlisted (CLAUDE_CONFIG_DIR
wins for claude; codex/gemini need ~/.codex|~/.gemini auth; FS sandbox is
TODOS.md:454 scope; the hermetic-sentinel canary proves config isolation),
and PTY extraArgs --mcp-config is a deliberate caller opt-in like env overrides.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs: document hermetic-by-default E2E + eval:bg detached runs in CONTRIBUTING

The Testing & evals section now tells contributors that local E2E runners
spawn children through a sealed clean room (allowlist-scrubbed env, seeded
CLAUDE_CONFIG_DIR, temp GSTACK_HOME, --strict-mcp-config) so local signal
matches CI, with EVALS_HERMETIC=0 as the escape hatch. The eval-tools list
gains the eval:bg* detached-run scripts (gstack-detach: SIGTERM-proof,
caffeinate-wrapped, machine-locked, run-scoped logs, EXIT= sentinel).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore: sync package.json to 1.58.1.0

The merge took main's package.json (1.58.0.0); gstack-version-bump repair
fixed the working tree but the change was left uncommitted. Without this the
committed tree disagrees with VERSION and CI's version-match test fails.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs: regenerate diagram SKILL.md with Conductor prose preamble

The diagram skill (new from main) was missing the Conductor-session prose
AskUserQuestion blocks that gen-skill-docs propagates to every SKILL.md.
Pure generated output; reproduced by bun run gen:skill-docs.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

CHANGELOG.md

@@ -1,5 +1,107 @@
 # Changelog
 
+## [1.58.1.0] - 2026-06-14
+
+## **Local evals stop lying. Spawned `claude` test children run in a sealed clean room,**
+## **and in Conductor every decision is a plain-text brief you answer with a letter.**
+
+Two things shipped here. First, the local E2E harness is now hermetic by default:
+every spawned agent (claude -p, the real-PTY plan-mode runner, the Agent SDK
+runner, plus the codex and gemini runners) gets an allowlist-scrubbed environment,
+a fresh seeded `CLAUDE_CONFIG_DIR`, a temp `GSTACK_HOME`, and `--strict-mcp-config`.
+Before this, a dev machine leaked the operator's `~/.claude` config, MCP servers
+(gbrain, Conductor), skills, `~/.gstack` decision logs, and `CONDUCTOR_*`/`CLAUDECODE`
+env into every child, so local eval results disagreed with CI for reasons that had
+nothing to do with the code under test. Now local signal matches CI. Set
+`EVALS_HERMETIC=0` to debug against real operator state.
+
+Second, in a Conductor session gstack no longer fights Conductor's flaky
+AskUserQuestion tool. It detects the session and renders every decision as a prose
+brief, a labeled question with a recommendation, per-option completeness scores, and
+"reply with a letter," enforced by a PreToolUse hook that denies the tool and
+redirects to prose. Destructive confirmations demand an explicit typed answer.
+
+Agents that launch long eval runs get `gstack-detach`: a SIGTERM-proof, idle-sleep-proof
+wrapper (fresh session + `caffeinate`) with a machine-wide lock so concurrent
+worktrees serialize instead of saturating the model API, run-scoped logs, and a
+guaranteed `EXIT=` sentinel so a poller never mistakes silence for success.
+
+### The numbers that matter
+
+Measured against the gate eval suite on a contaminated dev box (gbrain MCP up, live
+Conductor session, sibling worktrees). Reproduce: `bun test` (free unit + wiring
+tripwire) and `EVALS=1 EVALS_TIER=gate bun test test/skill-e2e-hermetic-canary.test.ts`.
+
+| Metric | Before | After | Δ |
+|--------|--------|-------|---|
+| Spawned-child env | full operator `process.env` | allowlist-scrubbed | sealed |
+| Runners hermeticized | 0 of 5 | 5 of 5 | +5 |
+| Operator MCP servers visible to child | all (gbrain, Conductor) | 0 (`--strict-mcp-config`) | isolated |
+| Config isolation proof | none | poisoned-operator sentinel canary | falsifiable |
+| Long eval runs surviving a turn-boundary SIGTERM | no | yes (`gstack-detach`) | survives |
+
+The clean room is falsifiable, not asserted: a `hermetic-sentinel` gate canary
+plants a poisoned operator config (a user `CLAUDE.md` + an MCP server) and fails if
+the child can see any of it, and a free static tripwire fails CI if any runner
+reverts to a raw `process.env` spread.
+
+### What this means for contributors
+
+Run evals locally and trust the result. You no longer have to push to CI to find
+out whether a failure was real or just your machine bleeding context into the agent.
+Three latent bugs the old harness hid surfaced the moment the suite ran clean and
+are fixed: a coverage-judge that scored carved skills against half a document, an
+ios-qa daemon test that collided on a shared pidfile under concurrency, and an
+operational-learning fixture missing a lib it imports. Start a run with
+`bun run eval:bg:gate`; flip `EVALS_HERMETIC=0` only when you deliberately want your
+real `~/.claude` in the loop.
+
+### Itemized changes
+
+#### Added
+- **Hermetic E2E environment** (`test/helpers/hermetic-env.ts`): allowlist env
+  builder (process basics, network/proxy vars, named `ANTHROPIC_*` auth, per-runner
+  `extraAllow`), pure `promotedEnv()` shared with `lib/conductor-env-shim.ts`, a
+  sync-memoized singleton temp dir (`<runRoot>/.claude` keeps the plan-file path
+  contract), a seeded `.claude.json` for non-interactive first run, and pid-aware GC
+  of crashed runs. Default-on; `EVALS_HERMETIC=0` restores the legacy env AND drops
+  `--strict-mcp-config`.
+- **Two gate-tier isolation canaries** (`test/skill-e2e-hermetic-canary.test.ts`):
+  `hermetic-canary` asserts env redirect + scrub + zero MCP servers + nonzero
+  API-key cost from the Bash tool_result (not model prose); `hermetic-sentinel`
+  proves the child cannot see a planted poisoned operator config.
+- **Static wiring tripwire** (`test/hermetic-wiring.test.ts`): free-tier invariants
+  that fail CI if any of the five runners drops `hermeticChildEnv()`, the gated
+  `--strict-mcp-config`, or leaks `process.env` through a callsite override.
+- **`gstack-detach`** + `eval:bg` / `eval:bg:all` / `eval:bg:gate` / `eval:bg:periodic`
+  scripts: detached, SIGTERM-proof, `caffeinate`-wrapped eval runs with a machine-wide
+  lock, per-run logs under `~/.gstack-dev/eval-runs/`, a watchdog, and an `EXIT=`
+  sentinel.
+- **Conductor prose AskUserQuestion**: when a Conductor session is detected, every
+  decision renders as a prose brief (labeled question, recommendation, per-option
+  completeness, reply-with-a-letter), enforced by a PreToolUse hook that denies the
+  tool and redirects. Auto-decide preferences still apply first; destructive
+  confirmations require an explicit typed answer. Installed for Conductor even in
+  non-interactive setup, with an upgrade migration for existing installs.
+
+#### Changed
+- All five E2E runners (`session-runner`, `claude-pty-runner`, `agent-sdk-runner`,
+  `codex-session-runner`, `gemini-session-runner`) spawn children through
+  `hermeticChildEnv()`. The Agent SDK runner now receives a COMPLETE hermetic env
+  via `Options.env` (the old "never pass env: to the SDK" rule was partial-env
+  replacement; a complete env is safe).
+- `hermetic-env.ts` is a global touchfile, so any change to it selects every E2E +
+  judge test.
+- CLAUDE.md documents hermetic-by-default local evals and retires the stale SDK env
+  warning.
+
+#### Fixed
+- The workflow LLM-judge now re-appends body-carved `sections/*.md` after the marker
+  slice, so carved skills (document-release) are judged on the full workflow the
+  agent executes instead of a half-document.
+- ios-qa daemon scenarios use unique pidfiles, fixing `already_running` collisions
+  under `bun test --concurrent`.
+
 ## [1.58.0.0] - 2026-06-12
 
 ## **Your documents grow diagrams. Mermaid and excalidraw fences render as real pictures,**