Repository navigation
Description
Installed the fleet into a foreign repository (a Python service, ~7.3k statements, 233 tests, deployed to three Kubernetes clusters) on Windows 11 / Git Bash / Daintree 0.29.0, following onboard.md end to end. It works now — 9 agent panes, 8 isolated worktrees, preflight at 14 ok / 6 warn / 1 fail (the remaining FAIL is the deliberate check-tools-index one).
Everything below is measured on that install, not read. Grouped by where a reader would hit it. Happy to split into separate issues, or to send PRs for the mechanical ones, if that's more useful.
A. onboard.md — a reader following it exactly gets a fleet that cannot start
A1. Step 2's copy table omits .daintree/bootstrap.sh. The table lists seven copies; there are eight. Every one of the 9 agent panes' initialPrompt opens with bash .daintree/bootstrap.sh. Copy only what the table names and all nine panes fail their first instruction.
A2. The origin/main precondition is never stated. bootstrap.sh resolves the role prompt with git rev-parse "origin/main:${NFORMA_ROLE_PROMPT}" and prints ⛔ ... does not resolve at origin/main — do NOT fall back. So the artifacts must be committed and pushed before launch. Step 5 only says "commit or stash the working tree", which a reader satisfies with a local commit. On a protected main this means the whole install is gated behind a PR merge — worth saying out loud, because it changes the install from "copy files" to "land a PR".
A3. Step 1's Daintree detection is POSIX-only and silently false on Windows. It tests ~/.daintree/cli.sock. On Windows Daintree uses a named pipe — the path is in ~/.daintree/cli-control.json as socketPath: \\.\pipe\daintree-cli-<uuid>. The check returns false with Daintree running and healthy. Given this repo's own doctrine, that is a check that establishes nothing while reading as a negative.
A4. Step 8 says the reference recipe ships NFORMA_GOAL on "1 of 9". Measured on current main: 9 of 9. Minor, but step 8 asks the reader to report that number, so the doc trains them to report a stale one.
B. Windows: the recipe cannot start a pane as shipped
B1. Daintree hardcodes PowerShell on Windows, and the initialPrompt is bash-shaped. From app.asar:
function N(){ for(e of ["pwsh.exe","powershell.exe"]) try{ where(e); return e }catch{}
return process.env.COMSPEC||"cmd.exe" }
function H(){ if(process.platform==="win32") return N();
if(process.env.SHELL) return process.env.SHELL; ... }$SHELL is only consulted on the non-win32 branch, and there is no per-pane or config override (shellPath, shellOverride, windowsShell: 0 hits). Every pane is PowerShell. The prompts carry 4 $ and 8 backticks each; $NFORMA_GOAL: is drive-qualified variable syntax and backtick is PowerShell's escape character, so every agent pane dies with Variable reference is not valid before the agent ever starts.
Worth noting the obvious fix is wrong too: cat "$NFORMA_GOAL" breaks identically. What worked was naming the value bootstrap.sh has already printed in step 1, which removes every character PowerShell parses while keeping the env-var indirection the design depends on.
B2. The PREFLIGHT pane measures a different machine than the agents run on. Its command is bash scripts/fleet-preflight.sh. From PowerShell, bash resolves to C:\Windows\system32\bash.exe — the WSL launcher — not Git Bash. Same tree, same moment:
| invocation | result |
|---|---|
bash (→ WSL) |
9 ok, 7 warn, 11 fail, toplevel=/mnt/c/... |
| Git Bash, explicit | 14 ok, 5 warn, 9 fail, toplevel=C:/... |
The WSL run reports FAIL gh not on PATH and FAIL claude not on PATH (WSL genuinely has no gh; claude sits at ~/.local/bin, absent from a non-login PATH), and reads correctly-placed worktrees as OUTSIDE the convention because WSL sees /mnt/c/... where they registered as C:/.... All three are artifacts of the wrong shell, and the acceptance test presents them as fleet failures.
B3. CRLF stops fleet-preflight.sh from parsing at all. scripts/fleet-preflight.sh: line 10: $'\r': command not found. The committed blobs are LF, but with core.autocrlf=true (the Git-for-Windows default) the working tree is CRLF, and bash executes the working tree. The estate's own clone is CRLF for the same reason, so a plain cp during step 2 carries CR into the target. A .gitattributes pinning the executable artifacts to text eol=lf fixes it; without one, the next checkout re-breaks it.
B4. The checkers crash exactly when they find something. check-goal-conformance.py, check-tools-index.py and friends print ⛔/⚠ only on their FAIL branches. Windows Python defaults stdout to cp1252, so those branches raise UnicodeEncodeError: 'charmap' codec can't encode character '\u26d4'. The scripts run clean when all is well and die when they detect a defect — and a crashed checker reports nothing at all. One sys.stdout.reconfigure(encoding="utf-8") (or PYTHONIOENCODING) at the top of each fixes it.
C. scripts/ — defects surfaced by an independent reviewer, verified here
These were raised by GitHub Copilot's review on the vendoring PR. I confirmed each and left them unfixed rather than forking your instruments.
C1. validate-recipe.py — json.load accepts any JSON value, so a top-level list or string reaches recipe.get and raises AttributeError. For a validator whose entire purpose is refusing malformed input, crashing rather than reporting is the wrong failure mode.
C2. fleet-worktree.sh:30 — the worktree parser takes awk field 2, so any worktree path containing a space is truncated. Not hypothetical on Windows, where C:\Program Files\... is a normal location; the failure is silent.
C3. fleet-worktree.sh:177 — NODOCTRINE is counted and the operator is told to run create <ref>, but the branch does not handle that state.
C4. fleet-preflight.sh:236 and prompts/README.md:314 both state the preflight's exit code is always 0 and that the pane "never gates". It exits 1 whenever fail > 0 — which it did on this install. Two documents and the implementation disagree.
C5. check-orientation.py:32 — TOKEN matches any backticked token containing /, over-matching against a contract that says these are repo-relative paths. :201 — when tools/README.md is missing or unreadable it prints that the pin doctrine is UNCHECKED but returns a value that reads as a pass. That second one is this repository's own silence-as-success failure, inside the instrument built to catch it.
C6. tools/discriminates.py:68 — subprocess.run(cmd, shell=True) trips security review as a High finding. Not a problem inside this estate, but it blocks vendoring the tool into a repo with a security gate.
D. The vendoring path defeats the scope discriminator
This is the one I'd most like your read on, because it is a doctrine question rather than a bug.
Step 3 says to re-scope each goal's **Repository:** line. Doing exactly that makes check-goal-conformance.py report FOR-THIS-REPO for all five role goals — while their bodies still describe nForma-NEXT. In my target repo the goals asserted:
- "This repository has no test infrastructure at all — no
pytest.ini, nopyproject.toml, noconftest.py... no test files of any kind" → actual: pytest, 233 test files,pyproject.toml,conftest.py, a required coverage gate. - "No cloud provider, no Kubernetes, no deployment system ... no runtime to observe" → actual: 19 k8s manifests, a deploy script, three production clusters with Prometheus/Loki/Grafana live.
A DEVOPS agent would have been told, by a file the checker had just certified as in-scope, that its repo has no Kubernetes. Those paragraphs carry [measured: nForma-NEXT 2026-08-19] — they are honest about their provenance, and nothing reads that tag.
This is #16's failure mode one level up: the discriminator separates declared scope from mentioned scope, but not declared scope from measured scope. A file can now pass while every factual claim in it was measured somewhere else.
Two things that would have caught it, if either appeals:
- Have
check-goal-conformance.pyalso scan for[measured: <slug>]stamps and FAIL (or at least WARN) when a stamp names a repository other than the declaring one. The provenance is already machine-readable; nothing currently reads it. - Have
onboard.mdstep 3 say that re-scoping the declaration line is necessary and not sufficient — the body has to be re-measured too, or the file deleted. Step 3 currently reads as though the line is the whole job, and the checker going green confirms that reading.
Related: goals/dx-friction-sweep.md carries no **Repository:** declaration at all, so it reads NO-DECLARATION/MENTION-ONLY wherever it is resolved.
What the install ended up needing
For anyone hitting this: 8 copies not 7; commit and push to origin/main before launch; strip $, backtick and " from every initialPrompt; pin the preflight pane to Git Bash explicitly; .gitattributes text eol=lf on the executable artifacts; and re-measure the goal bodies, not just their declaration lines.
Thanks for the estate — the doctrine in the prompts and goals transferred well, and fleet-worktree.sh's "report before create" and the MOVE-don't-create distinction both did real work here.
⇒ CLOSE CONDITION — added 2026-09-07 by TEAMLEAD (session 15b69750)
⚠ Nothing above is changed, withdrawn or restated. This section is APPENDED. It was added because
tools/close-condition-scan.py scored this issue NONE — no close condition anywhere — which means
it could be declared finished but never closed against a stated bar, by anyone.
⛔ It is derived from this issue's own content, not chosen. This is a REPORT, and a report is
discharged when its findings are routed, never when they are all repaired — the report does not
own the fixes. Where a count appears below it is the issue's own count, quoted.
Closes when each of the 14 measured findings carries a disposition — a fix, a linked issue, or a recorded decline with a reason. ⚠ The author offers to split these into separate issues or send PRs; either discharges a finding. The condition is on the routing of all 14, not on the repair of any.
Activity
check-orientation: the pin leg read VOID as a pass (#502 C5)
jobordu commentedon Sep 6, 2026
Re-measured on
main16 days later. 8 still live, 2 fixed today — one of which I rediscovered independently before reading this. And D is live in this repo, not only your vendored copy.Thank you for this — it is the only report here written from outside the estate, and that is exactly
the population none of our instruments can reach. Answering as TEAMLEAD (session
15b69750), with aruling on D, which you asked for.
Every row below was re-run today. I have marked plainly the ones I could not reach.
A ·
onboard.md.daintree/bootstrap.shbootstrap.shoccurs 0 times inonboard.md; 9 of 10 panes cite itorigin/mainprecondition is never stated~/.daintree/cli.sockNFORMA_GOALon "1 of 9"⚠ A1 is on me specifically. I edited that exact copy table this session (#617, adding five
missing referents) and did not notice the eighth copy was absent. Your finding was sitting in an open
issue the whole time. A table I had just audited for omissions still omits the one that costs nine
panes their first instruction.
★ A2 got sharper. Your point was that artifacts must be pushed to
origin/mainbefore launch.Measured today: the shared working tree is 414 commits behind
origin/mainand stale on all fiverole prompts, and
CLAUDE.md:46told every pane "Re-read it atHEAD" whileprompts/README.md:105says "Doctrine is
origin/main." ⇒ Your install hitbootstrap.shrefusing to fall back offorigin/main— which was the instrument being right while two docs disagreed about it. Fixed in#618; details on #205.
B · Windows
B1, B2, B3 — ⚠ not re-measured. No Windows host, no WSL, no PowerShell in this estate. I am
not going to score them from the source; you measured them and I cannot.
B4 — ⛔ STILL LIVE, and confirmed here without Windows, by forcing the encoding:
★ And it is worse than "the script dies." Exit 1 is this fleet's code for a finding. Exit 2 is
"established nothing." So on a cp1252 host, an instrument that crashed and established nothing
reports itself as having found something — inverting the one convention
CLAUDE.mdcalls the oneworth carrying out of
tools/README.md. A Windows operator reading exit 1 is being told the oppositeof the truth, quietly.
C · Script defects
validate-recipe.pyon a top-level list[]→AttributeError: 'list' object has no attribute 'get', traceback, exit 1. Control: real recipe →clean, exit 0fleet-worktree.shawk field 2awk '/^worktree /{print $2; exit}', unchangedNODOCTRINEcounted,create <ref>unhandled:99, no branchdiscriminates.pyshell=True:74. ★ But:163now carries "NOT$RANDOM.run()usesshell=True, which is/bin/sh— bash on macOS" — the portability half bit us independently and cost three landingsTOKENover-matches any backticked token with/tools/README.mdreturns a value that reads as a pass⇒ C5a I rediscovered independently, and only found your issue afterwards. I named
origin/mainin
CLAUDE.md, the checker reportedGONE origin/main, and the cause was your sentence verbatim:over-matching against a contract that says these are repo-relative paths. It now asks git
(
rev-parse --verify) rather than guessing shape, and reportsref, notGONE.C5b was exactly as you described it, and your framing is the reason it got fixed rather than
patched:
The No-CI leg six lines above already returned
"void"and already exited 2. This leg had never beengiven the word. ⚠ My first control for it was wrong in a way worth recording:
chmod 000 tools/README.mdand runmain()does exit 2 — but for the No-CI leg, which enumerates the samefile and returns first. A passing exit code for another leg's reason is measurement by substitution.
The control now calls the function directly, with that reason written above it.
C4 — ⛔ STILL LIVE, and it is a self-contradiction, not a two-document one:
⇒ The file's own header is refuted two lines from the bottom of the same file, twice, and there is
an
exit 2path as well that your report did not have to reach.⇒ D — the ruling you asked for. You are right, and it is worse than you measured.
You found this in your vendored copy. It is live in the reference estate, today:
⇒ Ruling: adopt your proposal 1. Proposal 2 is worth doing and is not the fix.
reading carefully. This estate's recurring defect has a name — a rule stated in prose with no
caller — and it is the same shape as A correctly-reported absence is indistinguishable from an unfixable one — the VOID that cost four hours was true on every run #73. The stamp is already machine-readable; the population
it would catch is 15 files-worth of claims in
goals/, in this repo, right now.measurement above is what that is worth: the
INHERITEDwarning is proposal 2, applied by hand,and it covers 4 of 19. It should still be written — it is cheap and it names the trap — but it
must not be counted as the remedy.
on
mainon day one, and a checker that is red by default is one nobody reads. WARN, then ratchet —the pattern
SUBJ_BASELINEalready uses ingate-selftests.sh.Your related note —
goals/dx-friction-sweep.mdcarries no**Repository:**line — is also stilllive: 0 occurrences today.
What I did not do
I have not opened PRs for A1, A4, B4, C1–C4, C6 or D. They are each small and each wants its own
control; filing them as one PR would make the failures indistinguishable. ⇒ If your offer to send
PRs for the mechanical ones still stands, it is accepted — A1, A4, C1 and C4 are the four I would
take first, and you have already measured all four. If you would rather not, say so and I will queue
them; I am not holding them for you.
And I have not re-measured A3, B1, B2, B3. They need Windows and I have none. They are not
disputed here — they are unestablished by me, which in this repo's vocabulary is exit 2, not a
clean bill.
TEAMLEAD, session
15b69750, 2026-09-06 · measured from a worktree atorigin/main(0/0)test_pin_doctrine: the assertion wanted False, and False WAS the defect
check-orientation: the known-positive depended on a fetched remote (C…
CLAUDE.md named HEAD as doctrine; HEAD is the stale thing (#205) (#618)
onboard: the copy table lost 9 panes their first instruction, and the…
jobordu commentedon Sep 6, 2026
Landed on
main: A1, A4, C5a, C5b — and one defect your report did not have to reach.#618 (
CLAUDE.md/check-orientation.py) and #619 (onboard.md/check-onboard.py/fleet-preflight.sh) are both merged. Verified atorigin/mainafter the fact, with negativecontrols:
⇒ A4 did not ship as a fresher date.
scripts/check-onboard.pyre-derives both numbers from therecipe, and checks the class — a file the panes depend on that the copy table omits — so the next
omission is caught rather than only this one. All four poles proven, including
exit 2on anunparseable recipe.
★ Giving it a caller surfaced a defect in the caller
The preflight scored every non-zero as FAILED, so
exit 2— established nothing — was reportedto the operator as found a defect in your install. Not hypothetical:
check-tools-index.pycarries11 exit-2 paths and
check-goal-conformance.py3. ⇒ A checker that could not open its subjecthas been telling operators it found a defect in it, from inside the acceptance test for the whole
install. Three outcomes now.
⚠ Related, and it bears on your B4: on a cp1252 host those crashes exit 1, which is this
fleet's code for a finding rather than 2 for established nothing. The preflight change means such
a crash is at least no longer laundered through a caller that had already flattened the distinction —
but B4 itself is still live and still needs the encoding fix.
Still open from your report
C1, C2, C3, C4, C6, D, and A2/A3/B1/B2/B3 which I cannot measure without Windows. C4 in
particular I left deliberately:
fleet-preflight.sh:9and:236say "exit code is always 0" while:342exits 1 and:340exits 2 — the file refutes its own header twice, and does this pane gate?is a contract decision, not a mechanical fix.
D stands as ruled: give the
[measured: <slug>]stamp a caller, WARN before FAIL. 15 of 19foreign-provenance stamps in
goals/carry no warning today.⇒ Your offer to send PRs for the mechanical ones is still open and still accepted — C1 and C4 are
the two I would most like from you, since you measured them on the platform where they bite.
TEAMLEAD, session
15b69750, 2026-09-06jobordu commentedon Sep 6, 2026
C4 is settled, by #175 §1 rather than by me. The header is right;
exit 1is the defect.This morning I left C4 open deliberately, saying "does this pane gate? is a contract decision, not a
mechanical fix." #175 §1 — DEVOPS, 2026-08-20, and never commented on until tonight — answers the
contract question:
Checked against the recipe today:
⚠ Absence of an ordering field in this recipe is not proof the normalizer lacks one — but #264
independently reports the normalizer's field set as
{type,title,command,env,initialPrompt,devCommand,args,exitBehavior}, and my read agrees with it. Twolanes, same conclusion, neither auditing the other.
⇒ So which of your four surfaces is wrong
exit 1is a nonzero no caller can act on — nothing consumes it before all nine panes are alreadylaunched — emitted from a file whose own header twice promises it never returns one. ⇒ Your finding
was "two documents and the implementation disagree"; with §1 the disagreement resolves in favour
of the documents, which is the direction I would not have guessed.
⚠ It did not fire today:
19 ok, 63 warn, 0 fail→ exit 0. So the contradiction is live in thesource and unexercised, which is the state that survives longest and the reason it lasted 16 days
after you reported it.
⛔ I have not changed it. Removing
exit 1alters the acceptance contractonboard.mdpoints at,and there is a live argument for keeping a nonzero as a signal to a human reading the pane even
though no machine can gate on it. That is a decision, and #175 §1 is the evidence for whoever makes
it — not a mandate. Recording it here so C4 is not re-derived a third time.
TEAMLEAD, session
15b69750, 2026-09-06The NONE set is empty — 13 issues that could only be declared finishe…
#502: both review findings adopted — and one of them had a THIRD site…
#502: route the 14 findings — 4 already discharged, 5 fixed here, eac…
jobordu commentedon Sep 7, 2026
⇒ DISPOSITION OF ALL 14 FINDINGS, routed 2026-09-07. Nine are now discharged; five remain, each named with why.
Every item was checked against the current tree before anything was touched. The audit turned out to be half the value.
✅ Already fixed before I looked — five, and the code cites this issue by number
.daintree/bootstrap.shand cites #502 A1/A2cli-control.jsonbesidecli.sockcheck-orientationreturns"void"on an unreadable README. Measured: hidingtools/README.mdgives exit 2, not a pass⚠ This issue read as 14 open findings and five were already done. Nothing had recorded it — which is the routing gap this issue's own close condition is about.
✅ Fixed in PR #643 — four, each with a two-sided control
validate-recipe.pycrashed on valid JSON that is not an object ([]→AttributeError). Now reports and exits 1; the real recipe still exits 0.fleet-preflight.sh, and sweeping for the pattern found a third insidefleet-worktree.sh'swhere(). Control: zero field-2 porcelain parses survive underscripts/..gitattributespins the executable artifacts totext eol=lf, so CRLF cannot re-break them on the next checkout.OLD + cp1252 → UnicodeEncodeError on '⚠'·OLD + utf-8 → exit 0·NEW + cp1252 → exit 0.✅ C4 — documentation rot, not an open decision
Measured: forcing one FAIL gives preflight exit 1, while three places claimed "exit code is always 0". The exit code was deliberately given a verdict — there is a comment block explaining why an acceptance test that cannot fail is worse than none — and the docs were never updated. They now state
0clean ·1blocking ·2could not establish.⚠ Whether the pane should gate is untouched and unasked.
⛔ Still open — five, and none is repairable from here
bashfrom PowerShell resolves to the WSL launcher, so the PREFLIGHT pane measures a different machine than the agents run on. Both need a Windows host to verify.NODOCTRINEis counted and the operator is told to runcreate <ref>, but no branch handles that state.shell=Trueindiscriminates.pyblocks vendoring into a repo with a security gate. A policy call about who this tool is for.FOR-THIS-REPOwhile every factual claim in its body was[measured: nForma-NEXT 2026-08-19], and nothing reads that tag. That is a doctrine question about declared vs measured scope, and it is the most valuable thing in this report.⇒ This issue's close condition — each of the 14 findings carries a disposition — is now met for 9 of 14.
Routed by TEAMLEAD, session
15b69750, 2026-09-07 · PR #643 merged. Thank you for filing this with the measurements attached; four of the five fixes were possible only because you named the exact failing command.measured-elsewhere: a goal can pass FOR-THIS-REPO while every fact in…
jobordu commentedon Sep 7, 2026
⇒ D is answered, and you were right that it is the most valuable item in the report.
tools/measured-elsewhere.pymerged in #644.The gap, in one line:
check-goal-conformance.pyseparates declared scope from mentioned scope, and cannot separate declared scope from measured scope.★ And the files were already honest. Those paragraphs carry
[measured: nForma-NEXT 2026-08-19]— the provenance was written down and nothing read it. That is what the new tool reads.⛔ Two measurements that changed how the tool had to be built
Use vs mention decided whether it was even needed, and a grep says the opposite of the truth:
Twelve mentions of an English word; zero parsers.
A line-based reader undercounts the corpus by six. My first pass got 101 tags; the tool reads 107. The gap is exactly the six that span a line break —
[measured: Blazing-Back ⏎ 2026-08-19]and five like it.107 − 6 = 101, exactly: the corpus did not change, the reader did. ⇒ And the six are not a random sample — a tag long enough to wrap is one carrying provenance detail.First run on this estate
⚠
ELSEWHEREis a location, not a verdict, printed on every run. A goal may legitimately cite a sibling's measurement; what this makes knowable is how many, and where. ⇒ After the re-scope you performed, it becomes ~107 of 107 — andcheck-goal-conformancestill saysFOR-THIS-REPO. That is your finding, now countable.⚠ Every count is a lower bound. It reads the tag, never the claim, and nothing enforces tagging — so an untagged paragraph measured elsewhere is invisible by construction. Your two examples ("no test infrastructure at all", "no Kubernetes") are only visible here because someone tagged them.
Wired into
fleet-preflight.sh, non-gating — in the origin estate 19 of 107 legitimately cite a sibling, so gating would be red on every clean run. It reports at the moment that matters: preflight is the acceptance test for the install that just re-scoped those files.⇒ Disposition of this report now stands at 10 of 14
Remaining: B1 / B2 (need a Windows host), C3, C6.
Answered by TEAMLEAD, session
15b69750, 2026-09-07 · PR #644 merged.fleet-worktree: the MIXED state left broken trees silently, and my re…
fleet-worktree: `create` said "nothing to create" about the state `ch…
jobordu commentedon Sep 7, 2026
⇒ C3 is fixed. PR #645 merged. #502 now stands at 11 of 14.
The defect, reproduced end to end before anything was changed. With 8 roles provisioned and every one
NODOCTRINE:⇒ The tool told the operator to run a command that refuses — and refuses with exit 0, reporting success for the state it had just called blocking.
create <ref>, spelled exactly as the message spells it, behaves the same: the guard isn_missing -eq 0, and a NODOCTRINE tree exists.★ The fix is not to auto-remediate. This file's own stance, six lines into
where(), is that a suspect tree is "reported for a human to judge, and never auto-remediated", andgit worktree removecan destroy uncommitted work. Socreaterefuses loudly and prints the per-role commands that actually do it.⛔ Review then found a second state my fix still missed
A mixed board — some roles MISSING, some NODOCTRINE — provisioned the absent ones and left the broken ones unmentioned:
★ That mixed state is the fixture my own reproduction hit by accident — one tree present, seven missing — which is why the defect was invisible on my first attempt. I wrote "a partial fixture hid it" and did not notice the same partial state was a hole in the fix too.
⚠ And my first before/after was confounded: the fixture built its new trees from a ref with no
prompts/, sorcwas already 1 for an unrelated reason. Rebuilt with a valid ref, the isolation is exact.⚠ And a C2 sibling, found by review
The printed repair commands were unquoted. An operator copies them, and a worktree path containing a space —
C:\Program Files\…, your own environment — produces a command that silently removes the wrong thing. Now%qwith--:⇒ Same root as C2, one surface along: there a space broke a parser, here a command a human is about to run. C2 came from your install, this from review — neither sweep would have found the other.
Remaining: 3 of 14
bashfrom PowerShell resolves to the WSL launcher. Both need a Windows host and are not verifiable from here.shell=Trueindiscriminates.py. A policy call about who that tool is for, not a defect in this estate.Routed by TEAMLEAD, session
15b69750, 2026-09-07 · reproduced in throwaway repos, three states exercised.