fix(auth): require an explicit streams session URL

[rickylabs]

Summary

• netscript plugin auth session list no longer infers http://localhost:4437/auth/sessions.
• Callers must pass --stream-url; omitted input fails before the session HTTP port is called and directs users to run aspire describe streams --format Json, select the streams HTTP endpoint, and append /auth/sessions.
• Focused tests prove explicit URL forwarding and the fail-loud/no-adapter-call path.
• The streams manifest and official-copy 4437 values are preserved as required compatibility metadata. No schema/copy redesign is included.

Scope

• Archetype / area: Archetype 6 CLI consumer within the frozen Archetype 5 plugin leaf
• Closes auth: session list --stream-url default pins localhost:4437 which no longer exists post-#1211 #1243
• Only coordinator-authorized added surface: packages/cli/src/public/features/plugins/auth/auth-plugin-command_test.ts
• Direct-to-main milestone leaf; coordinator alone merges or publishes

Slices

• Harness/research bootstrap — e49948bbf
• Contract drift and pause evidence — 69aaeba2a, f3cf40909
• Explicit URL/fail-loud behavior and focused tests — 3d32e9ee2
• Isolated formatting slice — a21224586
• Committed changed-file reporter config — 6242edabc
• Durable non-expensive gate evidence — 98d5d9654

Validation

All command receipts attest 6242edabc3679173c841e2e167f7f5786819e720 under
.llm/runs/fix-legacy-port-pin-sweep--0.0.7-wave0/receipts/.

• Focused structured check: 2 files, 0 diagnostics
• Focused structured test: 11 passed, 0 failed
• Root lint/fmt receipts: PASS; changed-file CLI lint/fmt reports: 0 findings
• quality:gate: PASS
• arch:check: PASS
• CLI doc lint: 1 package / 3 entrypoints / 0 errors
• CLI JSR audit: PASS (0 FAIL; existing WARN-only findings)
• CLI package publish dry-run: PASS; no publication performed
• scaffold.runtime: intentionally NOT_RUN and no lease requested, per coordinator instruction

Harness

• Run dir: .llm/runs/fix-legacy-port-pin-sweep--0.0.7-wave0/
• Phase: implementation complete; draft remains open at status:impl
• PLAN-EVAL: N/A: the authorized explicit-URL remedy remained locked/mechanical
• Next mandatory gates: substantive topic-orchestrator Tier-A review, then a fresh opposite-family IMPL-EVAL
• No runtime resources were started; no ready transition, merge, or publication is authorized

Definition of Done

• Every filed 4437 occurrence is removed or classified with evidence
• Session listing cannot silently contact the legacy port and gives actionable endpoint guidance
• Structured non-expensive gates, JSR audit/publish dry-run, quality:gate, and arch:check pass
• scaffold.runtime explicitly withheld by coordinator; no lease requested
• Topic-orchestrator Tier-A review is complete — [TIER-A: PASS_TO_IMPL_EVAL] comment 5286347517 (2026-08-13T21:01:47Z); artifact .llm/runs/fix-legacy-port-pin-sweep--0.0.7-wave0/review-tier-a.md
• Separate opposite-family IMPL-EVAL passes — PASS at evaluated head e6ba15ec6414c0a42b1f9870791131162ea71c36; verdict commit a949a6cd1777b0d05b1a3b45143de15951aa6dc2; artifact .llm/runs/fix-legacy-port-pin-sweep--0.0.7-wave0/evaluate.md; IMPL-EVAL phase comment 2026-08-14T23:20:27Z; fresh native Claude Opus 5/low Remote Control session 8c47751a-6a30-4dab-b25c-dbafe9873455, separate from the Codex generator

[rickylabs]

[PHASE: PLAN]

Research locked the bounded explicit-URL remedy and dead-pin sweep.

Evidence

1. Live issue auth: session list --stream-url default pins localhost:4437 which no longer exists post-#1211 #1243 fetched at 2026-08-13T20:23:46.556Z; state and timestamp are preserved in research.md.
2. Baseline and origin/main both equal 01e0960494c95ce56eb35892c211a095eb13e6ed.
3. feat(mcp): harden Aspire CLI endpoint discovery #1206's established Aspire endpoint discovery is the @netscript/mcp service-endpoint directory. Wiring it crosses this leaf's frozen package/composition boundary; no duplicate parser will be introduced.
4. PLAN-EVAL: N/A: the remaining issue-authorized explicit-URL/fail-loud remedy is mechanical.
5. Bootstrap commit: e49948bbfc0cb2990e81946f7303a89160d62964.

Next

• Implement the three live-pin removals/contract change within the declared surfaces.
• Preserve the generated skill's historical 4437 incident evidence.
• Keep scaffold/Aspire/Docker blocked until the topic orchestrator grants the singleton lease.

[rickylabs]

[PHASE: PLAN] [VERDICT: CHANGES_REQUESTED]

Current-main contract tests exposed a frozen-surface rescope blocker before any source commit.

Findings

1. C1 — manifest fields are still contractual. officialSource.backgroundPort is required by the shared plugin schema; serviceEntrypoint / serviceConfigKey / servicePort are atomically validated. Official-copy compatibility still consumes the values. The attempted deletion made the streams manifest invalid and was restored.
2. C2 — the explicit-URL behavior needs an undeclared test surface. Two existing auth command tests invoke session list without --stream-url. The proposed fail-loud source diff is intentionally uncommitted because auth-plugin-command_test.ts is outside the leaf contract.
3. Structured reporter evidence: exit 1, 10 passed / 8 failed / 5 unique failure groups. No durable receipt was minted against uncommitted source.
4. Drift evidence commit: 69aaeba2a6601f4003145d67799967eab08dbdce.

Next

Topic orchestrator must choose one same-thread resume:

• Narrow (recommended): authorize packages/cli/src/public/features/plugins/auth/auth-plugin-command_test.ts; retain/classify the compatibility manifest pins as not mechanically removable in this leaf. PLAN-EVAL remains N/A.
• Broad: expand to the shared manifest schema and official-copy compatibility surfaces; require revised plan plus separate PLAN-EVAL before implementation.

The expensive-gate lease has not been requested because the leaf cannot reach a source commit until this choice is made.

[rickylabs]

[PHASE: PLAN]

Leaf paused at the frozen contract boundary.

• Cleanup/blocker artifact slice: f3cf4090993ec60b5a23a3669efa2825b01064a4
• Exact proposed auth-command diff is preserved in worklog.md; the uncommitted product-source edit was removed with apply_patch.
• Worktree and explicit remote branch are clean/current.
• No further gates ran and no expensive lease was requested.
• PR remains draft at status:plan.

Next

Release coordinator decision is required before any source work: issue a replacement leaf contract naming the necessary surfaces/remedy, or disposition #1243 outside this leaf. The topic orchestrator cannot authorize a rescope.

[rickylabs]

Coordinator contract decision: authorize the narrow addition of packages/cli/src/public/features/plugins/auth/auth-plugin-command_test.ts. Preserve the manifest/copy compatibility fields that structured validation proved are still required; do not broaden into schema/copy redesign. Finish the explicit --stream-url / fail-loud guidance slice, structured non-expensive gates, and stop at Tier-A + separate IMPL-EVAL handoff. No merge or publish authority.

[rickylabs]

[PHASE: IMPL]

The coordinator-authorized explicit URL/fail-loud slice is implemented and pushed.

Evidence

• Semantic commit: 3d32e9ee2ee37dc9cebfe645f93e3a4ea479c215
• Review-isolated formatting: a212245867b77ab8d40e7330b2b7cb7409781a90
• Receipt head: 6242edabc3679173c841e2e167f7f5786819e720
• Evidence commit: 98d5d9654
• Focused check: 0 diagnostics; focused tests: 11/11; changed-file lint/fmt: 0 findings
• quality:gate, arch:check, CLI doc lint, JSR audit, and CLI publish dry-run: PASS
• Manifest/copy 4437 values remain unchanged compatibility metadata
• scaffold.runtime/Aspire/Docker: NOT_RUN; no lease requested; no resources started

Handoff

PR remains draft at status:impl. Topic orchestrator: perform substantive Tier-A review, then launch a fresh opposite-family IMPL-EVAL in this same leaf flow. Do not mark ready, merge, or publish.

[rickylabs]

Exact-head hygiene correction

• Old head: 98d5d9654d00ca3e737d68cb2a68c2e0223f4c1e
• New head: 786c5e78513706889c48e53664ba1bea9b9a51ae
• Artifact-only change: removed the space-only line from the preserved fenced diff in worklog.md; no product, test, or source edits.
• git diff --check origin/main...HEAD — raw exit code 0; output empty.

PR remains draft with exactly status:impl. No evaluator or additional gate was launched.

[rickylabs]

[PHASE: REVIEW] [TIER-A: PASS_TO_IMPL_EVAL]

Substantive topic-supervisor review is recorded at .llm/runs/fix-legacy-port-pin-sweep--0.0.7-wave0/review-tier-a.md.

• Reviewed semantic commit: 3d32e9ee2
• Receipt subject: 6242edabc
• Hygiene fix: 786c5e785
• Independent structured test rerun: 11 passed, 0 failed
• Scope: auth command plus coordinator-authorized focused test only; manifest/copy compatibility pins unchanged
• Durable non-expensive receipts: PASS
• Round-one whitespace finding: resolved; final git diff --check PASS

Verdict permits only a fresh separate opposite-family IMPL-EVAL. PR remains draft at status:impl; no ready, merge, issue-close, publish, or expensive-gate authority is granted.

[rickylabs]

IMPL-EVAL — PASS

Field	Value
Phase	IMPL-EVAL (fresh separate formal evaluation)
Verdict	PASS
Evaluated head	e6ba15ec6414c0a42b1f9870791131162ea71c36
Immutable base	01e0960494c95ce56eb35892c211a095eb13e6ed
Evaluator commit	a949a6cd1777b0d05b1a3b45143de15951aa6dc2
Artifact	.llm/runs/fix-legacy-port-pin-sweep--0.0.7-wave0/evaluate.md

Remote-control identity

Field	Value
Requested route	native Claude Opus 5 / effort low (owner-authorized)
Observed route	native Claude Opus 5 / effort low (respawnFlags in job state.json)
Session ID	8c47751a-6a30-4dab-b25c-dbafe9873455
Bridge ID	cse_01LmSFUzxkHGuH98fiDhgHxH (/remote-control enabled)
PID	2464105
cwd	/home/codex/repos/netscript-007-leaf-legacy-port-pin-sweep

Implementation was Codex / GPT-5.6 Sol / low; this is a fresh opposite-family session that performed no implementation. supervisor.md still names Fable 5 / medium for this lane — superseded by the coordinator instruction. Fable 5 remains unassigned; no silent substitute.

Head resolution

PR head (gh pr view --json headRefOid), remote branch head (FETCH_HEAD), and local HEAD were resolved independently and all equal e6ba15ec6. The stale af3dca0f5 in the older leaf-local evaluate-prompt.md is superseded and was not used.

Verified

1. No silent localhost:4437 default — --stream-url declares no default:; the only surviving 4437 is inside the error string.
2. Fails before the session adapter — the guard precedes any dependencies.sessions.list reference, and the new test pins listCalls === 0 on rejection.
3. Actionable discovery guidance — names aspire describe streams --format Json plus the /auth/sessions suffix in both the option description and the error.
4. Manifest/copy 4437 preserved as coordinator-classified compatibility metadata (issue comment 2026-08-13T20:33:34Z); neither file is in the diff. drift.md correctly explains why removal is not mechanical.
5. Formatting isolated — behavior in 3d32e9ee2, formatting in a21224586; diffed with -w --ignore-blank-lines the latter is pure re-wrapping and quote normalization.
6. Evidence honest — all 8 receipts PASS / exit 0 / gitHead == actualGitHead == 6242edabc, and the delta from that head to e6ba15ec6 touches only .llm/ artifacts, so no product file changed after the receipts. JSR report: 0 failing findings. Publish was dry-run only. deno.lock untouched.

Evaluator-executed checks at the evaluated head: focused auth suite 11 passed / 0 failed (reproduces the receipt claim exactly), deno check clean, git diff --check empty. No scaffold.runtime, Aspire, Docker, or publish was run or requested.

Non-blocking observations

• N1 — session revoke --auth-url still defaults to http://localhost:8094/api/v1/auth, the same pin class in the same file. Out of scope for auth: session list --stream-url default pins localhost:4437 which no longer exists post-#1211 #1243 and correctly left alone; recommend a follow-up issue.
• N2 — packages/cli/ is excluded from root deno fmt, so the reformat was elective rather than gate-enforced; the cli-fmt report passed via explicit file selection that bypasses the directory exclude. Acceptable as landed.
• N3 — plan.md says the omitted URL "fails at option parsing"; it is an action-time guard. Deliberate, so the Aspire guidance replaces a generic Cliffy error — wording nit only, the required property holds.

Issue state

#1243 item 1 is resolved; items 2 and 3 were reclassified as required compatibility metadata after structured validation disproved the filing assumption. The PR correctly references #1243 with no closing keyword. Whether that reclassification suffices to close #1243 is the coordinator's call and is not decided here.

Stop state

PR remains draft at status:impl, milestone 0.0.7. No merge, ready transition, publication, relabeling, issue mutation, or central state change was performed. This verdict authorizes none of them.

[rickylabs]

[PHASE: REVIEW] [TOPIC-RECONCILE: IMPL-EVAL PASS ACCEPTED] [TRANSITION: WITHHELD]

Topic orchestrator topic-fixes-0.0.7 (native Claude Opus 5/high, Remote Control) reconciled the order-2 gate.

Accepted

Field	Value
Verdict	IMPL-EVAL PASS
Evaluated source head	e6ba15ec6414c0a42b1f9870791131162ea71c36
Evaluator commit	a949a6cd1777b0d05b1a3b45143de15951aa6dc2 (only evaluate.md, +161)
Evaluator session	8c47751a-6a30-4dab-b25c-dbafe9873455 — native Claude Opus 5 / low / Remote Control, fresh, separate from the Codex generator 019ffcca-8bdc-7fb3-98c5-df90e2ae3b1f
Prior Tier-A	comment 5286347517 ([TIER-A: PASS_TO_IMPL_EVAL])

Definition of Done is now complete: both remaining boxes checked with linked evidence.

The verdict was not accepted on its headline. Six checkable claims were re-derived independently and all six hold: evaluated head equals the verdict commit's parent; the verdict commit touches only evaluate.md; the product delta over base 01e096049 is exactly auth-plugin-command.ts + auth-plugin-command_test.ts; deno.lock untouched; no product file changed between receipt head 6242edabc and the evaluated head, so the receipts still describe the evaluated tree; and --stream-url declares no default: with the guard preceding dependencies.sessions.list.

Transition withheld — not this lane's authority

This PR stays draft at status:impl. The topic orchestrator is barred from marking ready, relabeling, merging, publishing, or closing issues; those are coordinator-only. Three substantive reasons beyond the rule:

1. draft→ready would dispatch a rival evaluator. The transition auto-triggers one OpenHands IMPL-EVAL and enters status:impl-eval unless impl-eval:skip is applied first — a second evaluator against a head that already carries a fresh native opposite-family PASS, and applying that label is itself a relabel this lane cannot make.
2. Verdict head ≠ current PR head. PASS was rendered at e6ba15ec6; the PR head is now a949a6cd1. The only delta is the verdict artifact itself, which is the expected pattern — but any status:ready-merge close-gate/mirror run would evaluate at a949a6cd1, so the distinction should be explicit rather than inferred.
3. auth: session list --stream-url default pins localhost:4437 which no longer exists post-#1211 #1243 disposition is undecided. The body deliberately carries no closing keyword, so merging would not close auth: session list --stream-url default pins localhost:4437 which no longer exists post-#1211 #1243. Whether the coordinator's compatibility-metadata reclassification of the manifest/copy 4437 fields satisfies the issue is the coordinator's call; this lane does not decide it.

Open follow-up (not filed by this lane)

Evaluator finding N1, confirmed against source: session revoke --auth-url still defaults to http://localhost:8094/api/v1/auth (auth-plugin-command.ts:138) — the same pin class as #1243, in the same file, genuinely outside its narrowing. It needs its own issue or it will be lost.

No merge, ready transition, relabel, issue mutation, publication, or expensive-gate lease was taken.

[rickylabs]

@openhands-agent model=openrouter/deepseek/deepseek-v4-flash-0731 output=pr-comment iterations=800 phase=impl head=a949a6cd1777b0d05b1a3b45143de15951aa6dc2

Trusted base SHA: dd47210
Evaluated head SHA: a949a6c

use harness

SKILL

• netscript-harness — apply the formal IMPL-EVAL protocol and verdict vocabulary.
• openhands-handoff — publish one machine-readable OpenHands verdict.
• netscript-tools — run the smallest decisive repository-native gates without mutating source.
• netscript-doctrine — apply package/plugin doctrine when the changed surface requires it.

Act as the formal IMPL-EVAL session for this pull request. Do not edit files, create commits, push,
or repair findings. The trigger metadata supplies the trusted base SHA and immutable head SHA: read
the evaluator protocol, verdict definitions, and selected profiles from that base commit, then
evaluate the PR body, linked issues, run artifacts, final diff, review threads, and architecture
debt at the immutable head. Verify the approved plan or recorded PLAN-EVAL: N/A, design
checkpoint, acceptance criteria, static/runtime/consumer gates, public surface, lock hygiene, and
false-done states. For documentation changes, also read every changed document fully and hand-test
representative executable claims.

Return concise, severity-ranked findings with exact evidence and required action. End with exactly
one supported verdict line using OPENHANDS_VERDICT: PASS, OPENHANDS_VERDICT: FAIL_FIX,
OPENHANDS_VERDICT: FAIL_RESCOPE, OPENHANDS_VERDICT: FAIL_DEBT, or
OPENHANDS_VERDICT: FAIL_PLAN. Write the same verdict to OPENHANDS_SUMMARY_PATH.

[rickylabs]

OpenHands Agent — Completed

OPENHANDS_VERDICT: PASS

Model: openrouter/deepseek/deepseek-v4-flash-0731
Provider: OPENROUTER
Job status: success
Verdict source: summary-file
Summary source: agent
Housekeeping: ack=success trace=success commit-artifacts=skipped replies=skipped draft-pr=skipped
Trace: .llm/tmp/run/openhands/pr-1643/run-31850530180-1 (in the Actions artifact)
Artifact: openhands-agent-31850530180-1

OpenHands IMPL-EVAL — PR #1643 — fix(auth): require an explicit streams session URL

Summary

Formal IMPL-EVAL session (separate from the Codex generator) for PR #1643 on branch
fix/legacy-port-pin-sweep, evaluated at immutable head a949a6cd1777b0d05b1a3b45143de15951aa6dc2.
The change removes the silent session list --stream-url default of
http://localhost:4437/auth/sessions; an omitted --stream-url now throws an actionable error
telling the operator to run aspire describe streams --format Json, append /auth/sessions to the
streams HTTP endpoint, and pass it explicitly — before the session HTTP adapter is ever called. The
change is narrow, pinned by focused tests, and stays within the coordinator-authorized surface.

The PR was already reviewed (Tier-A), and a prior separate opposite-family IMPL-EVAL recorded PASS
(evaluate.md, verdict commit a949a6cd). This session independently reproduces the verdict at the
same head.

Verdict: PASS

Changes

Product delta over the PR's true base (merge-base 01e09604) is exactly two files, both in the
authorized surface:

• packages/cli/src/public/features/plugins/auth/auth-plugin-command.ts — session list
--stream-url declared with no default; action throws the Aspire guidance error when omitted;
  guard precedes sessions.list so the adapter is unreachable on omission.
• packages/cli/src/public/features/plugins/auth/auth-plugin-command_test.ts — existing session list calls updated to pass --stream-url; new test session list fails loudly when the stream URL is omitted asserts both the rejection and listCalls === 0 (fail-before-adapter pinned).

Everything else in the 31-file PR diff is .llm/runs/fix-legacy-port-pin-sweep--0.0.7-wave0/ run
artifacts and receipts, plus the standalone cli-quality-deno.json lint/fmt config. deno.lock is
untouched. No packages//plugins/ file outside the two above was modified, added, or deleted.
The streams manifest and official-copy 4437 fields remain as coordinator-classified compatibility
metadata (correctly preserved).

Validation (independently rerun at head a949a6cd)

• deno check --unstable-kv <auth test> → clean, 0 diagnostics.
• deno test --allow-all --unstable-kv <auth test> → 11 passed, 0 failed (reproduces receipt).
• deno lint -c <run cli-quality-deno.json> <2 files> → 0 findings (root config intentionally
  excludes packages/cli; standalone config used, matching worklog).
• deno fmt -c <run cli-quality-deno.json> --check <2 files> → 0 findings.
• git diff --check over base..head → empty; worktree clean before/after validation.
• deno.lock unchanged by this evaluator (no cache --reload, no lock-write; git status clean).
• Confirmed --stream-url has no default:; the only surviving 4437 occurrence is the
  human-readable error string (line 117); guard at line 115 precedes sessions.list at line 122.
• No open review threads (0 PR review comments, reviewDecision empty); no unanswered findings.

Responses to review / issue comments

No open review threads or unanswered review comments exist on the PR. No response required.

Remaining risks (non-blocking)

• N1 (tracked follow-up, not a blocker): session revoke --auth-url retains a
  http://localhost:8094/api/v1/auth default — same defect class, but outside auth: session list --stream-url default pins localhost:4437 which no longer exists post-#1211 #1243/authorized
  scope. Already noted by the prior evaluator and Tier-A review; recommend a follow-up issue.
• N2 (config note): packages/cli/ is excluded from root deno fmt; the formatting sweep in
  a2122458 was elective and isolated in its own commit (behavior-preserving). Root fmt does not
  cover this path.
• Plan wording ("fails at option parsing") is slightly imprecise — it is an action-time guard — but
  the required fail-before-adapter property holds and is test-pinned; not blocking.
• scaffold.runtime was intentionally NOT_RUN / no lease requested. This is a narrow CLI
  option-contract change fully covered by focused unit tests; per plan/coordinator direction the
  withheld runtime gate is not evidence this verdict needed. PR stays ready/use-case; no merge or
  publication is authorized by this verdict.

Verdict

OPENHANDS_VERDICT: PASS

Run: https://github.com/rickylabs/netscript/actions/runs/31850530180