Creatidy/scarcity-router
5
0
Fork 0
Code Issues 13 Pull requests Projects Releases Packages Wiki Activity Actions

M07 Stage 2 reopen evaluation: all gates pass under single-owner scope — Stage 2 REOPENED (D-061, #92) #157

Merged
adrian.tkacz merged 3 commits from adapter/zcode-cli-stage2 into develop 2026-09-28 20:01:25 +00:00
Conversation 3 Commits 3 Files changed 5 +535 -9
kilo commented 2026-09-28 15:26:18 +00:00
Member
Copy link

Closes nothing — this is the evidence-preserving deliverable of the M07 Stage 2 reopen evaluation (issue #92, reopened). Decision record: D-061 in docs/decisions.md, revised 2026-09-28 on the owner's direction. Gates A, B and D pass (first evaluation, unchanged); Gate C re-evaluated: PASS, scoped to the recorded single-owner deployment scope. Stage 2 is REOPENED — implementation authorized subject to recorded constraints and retained exclusions. No adapter code is in this PR.

History of this PR

  • 112e950 — first evaluation: gates A/B/D PASS; gate C concluded UNCLEAR/BLOCKED from a Stage-1-style reading ("proxy-style use lacks vendor authorization").
  • The owner rejected that gate C reasoning as conflating Scarcity Router's architectural role with the Terms' "unauthorized proxy server" clause and directed a re-evaluation of gate C only, against the concrete deployment scope, using current official ZCode/Z.ai sources. Gates A/B/D are not re-litigated.
  • 8781060 + 3be4b6b — gate C re-evaluated clause-by-clause against the six official sources: PASS, scoped. D-061 revised; Stage 2 reopened through D-047's own point-4 reopen procedure. An independent adversarial review specifically of the gate C reasoning was run on the exact remediation head; its two merge blockers (leftover present-tense cancellation statements in docs/architecture.md and docs/providers.md) and four defers were remediated in 3be4b6b (review record in the comments).

Deployment scope the terms verdict is scoped to

One natural-person account owner; the owner's own paid GLM Coding Plan; ZCode installed and authenticated through supported mechanisms; Scarcity Router launches the official ZCode CLI locally as a subprocess; all credentials remain local to ZCode's own store; no third party receives account access; no public or multi-user gateway; no resale, no account pooling, no quota circumvention; no scraping, private IPC, reverse engineering or spoofing; ordinary coding work inside the owner's authorized workspaces. Any widening voids the evaluation (D-061 point 6).

Gate C re-evaluation (evidence doc §5) — the five questions

  1. Proxy server? No. The prohibition sits in ZCode ToS §IV's network-security list: "using ZCode as a virtual server, unauthorized proxy server, or mail server". The architecture has no proxied resource, no proxy beneficiary, no network/service boundary — the only model client is the owner's own authenticated CLI process launched locally; the router exposes no listener fronting ZCode and holds no credentials. The first pass's inference from the product's routing function is retracted in the record.
  2. Third party? No. The exclusivity clauses address other persons or organizations ("anyone other than the original registrant"; Usage Policy: "Account sharing or multi-user access is prohibited"; Subscription Terms §4.3: "bulk or automated usage on behalf of others"). ZCode ToS §III deems account activities "your actions" and §IV deems CLI operations "an extension of your own actions" — owner-controlled software is the owner's instrument, not another principal.
  3. Automation ban? No. The automated-means clauses are purpose-bound to scraping and unauthorized access (ZCode ToS §IV; Z.AI ToU §III.4(b): "for the purpose of accessing, obtaining, or monitoring any unauthorized data…"). The vendor's own headless --prompt path and automation surfaces confirm automation is a supported, regulated pattern (oversight duty; security-file clause → recorded as constraints), not a forbidden one. The S1/S2 wording asymmetry (missing "unauthorized" qualifier) is acknowledged and resolved by the clause chapeau.
  4. Commercial use? Satisfied. ZCode ToS §VI.2 lifts the non-commercial limitation when the user has "subscribed to the corresponding services on ZCode and paid the applicable fees". Official documentation (devpack overview; the vendor's "Using GLM Coding Plan in ZCode" guide with its "Use Subscription" binding step; ToS §IV/VIII paid-plan rules) establishes the paid GLM Coding Plan as that corresponding service for coding through ZCode. The Terms require no separate written vendor authorization for this path — the written-agreement reservation in Subscription Terms §4.2 governs use outside supported tools, which this architecture avoids by construction (the only model client is the official tool).
  5. Remaining prohibitions? None. Clause-by-clause sweep (§5.5), including the traffic-hijacking, competing-models, high-risk automated-decision and "damage our interests" catch-all clauses — none reaches the deployment. Sources, all retrieved 2026-09-28 and quoted verbatim in the evidence document: ZCode ToS (effective 2026-06-15), Z.AI platform ToU (2026-04-14), GLM Coding Plan Usage Policy, Subscription Terms §4, devpack overview, devpack ZCode guide.

Change (documentation only — five files)

  1. docs/zcode-adapter-stage2-evidence.md — §5 rewritten as the clause-by-clause re-evaluation with the scope statement and source list; §6 gate table (C: PASS scoped; E: AUTHORIZED, not entered here); §7 now the implementation constraints and retained exclusions.
  2. docs/decisions.md — D-061 revised: both D-047 reopen conjuncts met; Stage 2 REOPENED; D-047's NO-GO superseded through its own point-4 procedure; implementation authorized subject to constraints/exclusions.
  3. Dated M07 cross-references in docs/roadmap.md, docs/architecture.md (both tables and the dependency paragraph) and docs/providers.md (both paragraphs) — no present-tense "cancelled/NO-GO" statements remain outside historical narration.

Test/check results

Full gate on the exact head 3be4b6b: uv sync --only-dev OK; make check 2190 tests OK; uv run basedpyright 0 errors, 0 warnings, 0 notes; git diff --check clean. origin/develop..HEAD is exactly this issue's three commits, no merge commit.

UX impact

None — documentation only; no runtime, CLI, REST or MCP behavior changes. Documentation now states the current truth (Stage 2 reopened; implementation authorized under recorded constraints) instead of the stale "terms unresolved" state.

Security classification

Not security-critical: no code, credentials, transport or execution-surface changes. The evidence document records the credential-handling discipline and the constraints any future adapter must carry (explicit safe mode always passed; never extract or reuse credentials; never weaken ZCode security configuration; supported-tools-only quota use; operator oversight preserved).

Reviewer notes

  • The adapter implementation is a follow-up task; this PR contains no code.
  • Retained exclusions (D-061 point 6 = evidence §7): third-party credential/account sharing; public or multi-user gateway; resale/repackaging; account pooling; quota circumvention; use outside officially supported tools; providing model capabilities as a service to third parties; security-configuration bypass.
  • Historical records (D-047, Stage-1 evidence doc) remain untouched as history; D-061 now supersedes D-047's NO-GO via D-047 point 4.
  • Per repository governance: PR left open for human review; never merged by agents.
Closes nothing — this is the evidence-preserving deliverable of the M07 Stage 2 reopen evaluation (issue #92, reopened). Decision record: **D-061** in `docs/decisions.md`, revised 2026-09-28 on the owner's direction. **Gates A, B and D pass (first evaluation, unchanged); Gate C re-evaluated: PASS, scoped to the recorded single-owner deployment scope. Stage 2 is REOPENED — implementation authorized subject to recorded constraints and retained exclusions. No adapter code is in this PR.** ## History of this PR - `112e950` — first evaluation: gates A/B/D PASS; gate C concluded UNCLEAR/BLOCKED from a Stage-1-style reading ("proxy-style use lacks vendor authorization"). - The owner rejected that gate C reasoning as conflating Scarcity Router's architectural role with the Terms' "unauthorized proxy server" clause and directed a re-evaluation of gate C **only**, against the concrete deployment scope, using current official ZCode/Z.ai sources. Gates A/B/D are not re-litigated. - `8781060` + `3be4b6b` — gate C re-evaluated clause-by-clause against the six official sources: **PASS, scoped**. D-061 revised; Stage 2 reopened through D-047's own point-4 reopen procedure. An independent adversarial review specifically of the gate C reasoning was run on the exact remediation head; its two merge blockers (leftover present-tense cancellation statements in `docs/architecture.md` and `docs/providers.md`) and four defers were remediated in `3be4b6b` (review record in the comments). ## Deployment scope the terms verdict is scoped to One natural-person account owner; the owner's own paid GLM Coding Plan; ZCode installed and authenticated through supported mechanisms; Scarcity Router launches the official ZCode CLI locally as a subprocess; all credentials remain local to ZCode's own store; no third party receives account access; no public or multi-user gateway; no resale, no account pooling, no quota circumvention; no scraping, private IPC, reverse engineering or spoofing; ordinary coding work inside the owner's authorized workspaces. **Any widening voids the evaluation** (D-061 point 6). ## Gate C re-evaluation (evidence doc §5) — the five questions 1. **Proxy server? No.** The prohibition sits in ZCode ToS §IV's network-security list: "using ZCode as a virtual server, unauthorized proxy server, or mail server". The architecture has no proxied resource, no proxy beneficiary, no network/service boundary — the only model client is the owner's own authenticated CLI process launched locally; the router exposes no listener fronting ZCode and holds no credentials. The first pass's inference from the product's routing function is retracted in the record. 2. **Third party? No.** The exclusivity clauses address other persons or organizations ("anyone other than the original registrant"; Usage Policy: "Account sharing or multi-user access is prohibited"; Subscription Terms §4.3: "bulk or automated usage on behalf of others"). ZCode ToS §III deems account activities "your actions" and §IV deems CLI operations "an extension of your own actions" — owner-controlled software is the owner's instrument, not another principal. 3. **Automation ban? No.** The automated-means clauses are purpose-bound to scraping and unauthorized access (ZCode ToS §IV; Z.AI ToU §III.4(b): "for the purpose of accessing, obtaining, or monitoring any unauthorized data…"). The vendor's own headless `--prompt` path and automation surfaces confirm automation is a supported, regulated pattern (oversight duty; security-file clause → recorded as constraints), not a forbidden one. The S1/S2 wording asymmetry (missing "unauthorized" qualifier) is acknowledged and resolved by the clause chapeau. 4. **Commercial use? Satisfied.** ZCode ToS §VI.2 lifts the non-commercial limitation when the user has "subscribed to the corresponding services on ZCode and paid the applicable fees". Official documentation (devpack overview; the vendor's "Using GLM Coding Plan in ZCode" guide with its "Use Subscription" binding step; ToS §IV/VIII paid-plan rules) establishes the paid GLM Coding Plan as that corresponding service for coding through ZCode. The Terms require no separate written vendor authorization for this path — the written-agreement reservation in Subscription Terms §4.2 governs use outside supported tools, which this architecture avoids by construction (the only model client is the official tool). 5. **Remaining prohibitions? None.** Clause-by-clause sweep (§5.5), including the traffic-hijacking, competing-models, high-risk automated-decision and "damage our interests" catch-all clauses — none reaches the deployment. Sources, all retrieved 2026-09-28 and quoted verbatim in the evidence document: ZCode ToS (effective 2026-06-15), Z.AI platform ToU (2026-04-14), GLM Coding Plan Usage Policy, Subscription Terms §4, devpack overview, devpack ZCode guide. ## Change (documentation only — five files) 1. `docs/zcode-adapter-stage2-evidence.md` — §5 rewritten as the clause-by-clause re-evaluation with the scope statement and source list; §6 gate table (C: PASS scoped; E: AUTHORIZED, not entered here); §7 now the implementation constraints and retained exclusions. 2. `docs/decisions.md` — **D-061 revised**: both D-047 reopen conjuncts met; Stage 2 REOPENED; D-047's NO-GO superseded through its own point-4 procedure; implementation authorized subject to constraints/exclusions. 3. Dated M07 cross-references in `docs/roadmap.md`, `docs/architecture.md` (both tables and the dependency paragraph) and `docs/providers.md` (both paragraphs) — no present-tense "cancelled/NO-GO" statements remain outside historical narration. ## Test/check results Full gate on the exact head `3be4b6b`: `uv sync --only-dev` OK; `make check` **2190 tests OK**; `uv run basedpyright` 0 errors, 0 warnings, 0 notes; `git diff --check` clean. `origin/develop..HEAD` is exactly this issue's three commits, no merge commit. ## UX impact None — documentation only; no runtime, CLI, REST or MCP behavior changes. Documentation now states the current truth (Stage 2 reopened; implementation authorized under recorded constraints) instead of the stale "terms unresolved" state. ## Security classification Not security-critical: no code, credentials, transport or execution-surface changes. The evidence document records the credential-handling discipline and the constraints any future adapter must carry (explicit safe mode always passed; never extract or reuse credentials; never weaken ZCode security configuration; supported-tools-only quota use; operator oversight preserved). ## Reviewer notes - The adapter implementation is a follow-up task; this PR contains no code. - Retained exclusions (D-061 point 6 = evidence §7): third-party credential/account sharing; public or multi-user gateway; resale/repackaging; account pooling; quota circumvention; use outside officially supported tools; providing model capabilities as a service to third parties; security-configuration bypass. - Historical records (D-047, Stage-1 evidence doc) remain untouched as history; D-061 now supersedes D-047's NO-GO via D-047 point 4. - Per repository governance: PR left open for human review; never merged by agents.
docs: M07 Stage 2 reopen evaluation — official CLI verified, terms gate unresolved (D-061, #92)
All checks were successful
ci / check (pull_request) Successful in 7m57s
Details
112e950313 
Gate A (supported interface): PASS — ZCode v3.14.3 ships a first-party
Apache-2.0 CLI with a documented non-interactive path (--prompt, --mode,
strict --output-format text|json|stream-json, NDJSON events, doctor --json).
Gate B (headless permission model): PASS — --mode build|plan|edit with the
fail-closed DenyPermissionBroker; the yolo default is explicit and avoidable.
Gate C (terms): UNCLEAR/BLOCKED — terms unchanged since D-047's Stage-1
analysis; proxy-style automation still lacks vendor authorization.
Gate D (live spike): interface PASS (built from pinned source, probed in a
disposable workspace); completion environment-blocked (custom provider TLS).
Gate E: NOT ENTERED — Stage 2 remains NO-GO; D-047 not superseded.

Docs-only: new stage-2 evidence document, D-061 record, dated M07 row
cross-references in roadmap/architecture/providers.
fix: review remediation on #157 (D-061, #92)
All checks were successful
ci / check (pull_request) Successful in 8m3s
Details
87810603a5 
MERGE_BLOCKER: the evidence doc's second NOTICE.md quotation spliced a
lead-in that does not exist in NOTICE.md; replaced with the verbatim
subject ('子代理、工作流与后台任务', NOTICE.md:21) as an unquoted-paraphrase
lead, and aligned D-061's NOTICE characterization.

DEFERs folded in (same records, one-line each): unresolvable 'point-9'
pointer restated inline; gate-D framing harmonized between the evidence
doc and D-061; workspace version provenance corrected (zcode-cli 0.16.9
injected via __CLI_VERSION__, inner @zcode/cli 0.1.0); prompt-command.ts
shutdown-handler citation 142-151 -> 144-151.

Full gate re-run on this head: make check OK, basedpyright 0, git diff
--check clean.
kilo commented 2026-09-28 15:51:43 +00:00
Author
Member
Copy link

Independent exact-head review record

Reviewed head: 112e950 (single commit on develop @ 6fec46c). Reviewer: independent agent session (GLM-5.3-Flash), read-only, verified 20+ specific citations against the pinned first-party ZCode source tree (v3.14.3 @ 29628c9).

Verdict: CHANGES_REQUESTED — one MERGE_BLOCKER, four DEFER. UX impact: none (reviewer-assessed; docs-only, M07 rows became more informative).

MERGE_BLOCKER (finding 1): the evidence doc's second NOTICE.md quotation spliced a lead-in ("无人值守的自动化…") that does not exist in NOTICE.md; the verbatim fragment's actual subject is the automation-risk table row "子代理、工作流与后台任务" (subagents, workflows and background tasks). Since the entire value of this record is trustworthy provenance, a direct quotation had to be verbatim.

DEFER findings: (2) unresolvable "point-9" pointer to the agent task briefing; (3) gate-D framing tension between the evidence doc's gate table and D-061's "supporting evidence" phrasing; (4) version-provenance nit (0.16.9 is the zcode-cli workspace root version injected via __CLI_VERSION__, not the inner @zcode/cli manifest 0.1.0); (5) off-by-two citation prompt-command.ts:142-151 → 144-151.

Remediation (the single permitted round) — head now 8781060

8781060 fixes the MERGE_BLOCKER exactly as suggested (verbatim subject as an unquoted-paraphrase lead: ("子代理、工作流与后台任务") "可继续调用模型、执行命令并产生费用或外部副作用" — both strings machine-verified verbatim against NOTICE.md:21) and folds in all four DEFERs as one-line corrections in the same two records. No other files touched; origin/develop..HEAD remains merge-commit-free.

Final verification on 8781060: blocker confirmed fixed (spliced lead-in absent repo-wide; quotation verbatim-matched programmatically); diff is exactly the five docs files; full gate re-run: make check OK, uv run basedpyright 0 errors, git diff --check clean.

Per the bounded review lifecycle, review is closed: no further remediation rounds; final verification checked only the identified blocker and obvious regressions. Human merge gate remains open — not merged, develop untouched.

## Independent exact-head review record **Reviewed head:** `112e950` (single commit on `develop` @ `6fec46c`). Reviewer: independent agent session (GLM-5.3-Flash), read-only, verified 20+ specific citations against the pinned first-party ZCode source tree (v3.14.3 @ `29628c9`). **Verdict: CHANGES_REQUESTED** — one MERGE_BLOCKER, four DEFER. `UX impact: none` (reviewer-assessed; docs-only, M07 rows became more informative). **MERGE_BLOCKER (finding 1):** the evidence doc's second `NOTICE.md` quotation spliced a lead-in ("无人值守的自动化…") that does not exist in NOTICE.md; the verbatim fragment's actual subject is the automation-risk table row "子代理、工作流与后台任务" (subagents, workflows and background tasks). Since the entire value of this record is trustworthy provenance, a direct quotation had to be verbatim. **DEFER findings:** (2) unresolvable "point-9" pointer to the agent task briefing; (3) gate-D framing tension between the evidence doc's gate table and D-061's "supporting evidence" phrasing; (4) version-provenance nit (`0.16.9` is the `zcode-cli` workspace root version injected via `__CLI_VERSION__`, not the inner `@zcode/cli` manifest `0.1.0`); (5) off-by-two citation `prompt-command.ts:142-151` → `144-151`. ## Remediation (the single permitted round) — head now `8781060` `8781060` fixes the MERGE_BLOCKER exactly as suggested (verbatim subject as an unquoted-paraphrase lead: `("子代理、工作流与后台任务") "可继续调用模型、执行命令并产生费用或外部副作用"` — both strings machine-verified verbatim against `NOTICE.md:21`) and folds in all four DEFERs as one-line corrections in the same two records. No other files touched; `origin/develop..HEAD` remains merge-commit-free. **Final verification on `8781060`:** blocker confirmed fixed (spliced lead-in absent repo-wide; quotation verbatim-matched programmatically); diff is exactly the five docs files; full gate re-run: `make check` OK, `uv run basedpyright` 0 errors, `git diff --check` clean. Per the bounded review lifecycle, review is closed: no further remediation rounds; final verification checked only the identified blocker and obvious regressions. Human merge gate remains open — not merged, `develop` untouched.
kilo changed title from M07 Stage 2 reopen evaluation: official CLI verified, terms gate unresolved — Stage 2 remains NO-GO (D-061, #92) to M07 Stage 2 reopen evaluation: all gates pass under single-owner scope — Stage 2 REOPENED (D-061, #92) 2026-09-28 17:14:15 +00:00
kilo commented 2026-09-28 17:14:39 +00:00
Author
Member
Copy link

Independent adversarial review — round 2 (Gate C re-evaluation), exact reviewed head 3be4b6b837141dfa5456719d9380cf75212f63f2

Scope frozen per governance: the Gate C reasoning only (evidence doc §5–§7, D-061, cross-references) against the six official source texts retrieved 2026-09-28. Gates A/B/D not re-litigated. One initial review, one remediation, final verification — bounded lifecycle applied.

Verdict: CHANGES_REQUESTED (2 MERGE_BLOCKER, 4 DEFER). The Gate C reasoning itself was confirmed on every dimension the directive set:

  • Quote accuracy: PASS. All 24 quoted fragments in evidence §5 and D-061 verified character-exact against the source files (internal ellipses correctly marked); section attributions verified against the retrieved page structure (§III/§IV/§VI/§VIII offsets).
  • Missed clauses: none that clearly prohibits. The reviewer's independent sweep of all sources found no clause reaching the single-owner local-subprocess scope.
  • Soundness: PASS. The retraction of the first pass's function-based "proxy-style use" inference is justified by clause text, the §IV security chapeau, and the Terms' own deeming language ("deemed your actions", "extension of your own actions").
  • Classification: PASS is correct under the owner's decision rule — no applicable clause prohibits the exact deployment; the two interpretive readings are supported by clause objects, not left genuinely ambiguous; exclusions retained.

Blockers found and fixed in 3be4b6b (both were update-discipline slips, not reasoning errors):

  1. docs/architecture.md dependency paragraph still said "Stage 2 cancelled by D-047" in present tense → updated to "reopened by D-061".
  2. docs/providers.md Z.ai-preset paragraph still asserted "the M07 NO-GO does not remove Z.ai HTTP/API support" in present tense → recast as "D-047's since-superseded M07 NO-GO never removed…".

Defers folded into the same remediation:

  • §5.3 now acknowledges the S1/S2 wording asymmetry (S1 lacks the "unauthorized" qualifier) and resolves it by the clause chapeau plus the vendor's own automation surfaces.
  • §5.5 now name-checks the four additional swept clauses (traffic/engagement, competing-models, high-risk automated-decision, S2 Additional Terms for API Services) each with a not-reached rationale.
  • Evidence §7 exclusion list harmonized with D-061 point 6 (8 items, identical substance).
  • The S1 quote-verification extraction artifact was regenerated; every quoted fragment now verifies against the retrieval files.

Final verification on 3be4b6b: uv sync --only-dev OK; make check 2190 tests OK; uv run basedpyright 0 errors, 0 warnings, 0 notes; git diff --check clean. origin/develop..HEAD = exactly this issue's three commits (112e950, 8781060, 3be4b6b), no merge commit, all touched files in scope (docs only).

UX impact: none — documentation only.

Reviewer model: session model (GLM-5.3-Flash); not security-critical classification, so no additional security gate applies. PR remains open, unmerged, for human review.

**Independent adversarial review — round 2 (Gate C re-evaluation), exact reviewed head `3be4b6b837141dfa5456719d9380cf75212f63f2`** Scope frozen per governance: the Gate C reasoning only (evidence doc §5–§7, D-061, cross-references) against the six official source texts retrieved 2026-09-28. Gates A/B/D not re-litigated. One initial review, one remediation, final verification — bounded lifecycle applied. **Verdict: CHANGES_REQUESTED** (2 MERGE_BLOCKER, 4 DEFER). The Gate C reasoning itself was confirmed on every dimension the directive set: - **Quote accuracy: PASS.** All 24 quoted fragments in evidence §5 and D-061 verified character-exact against the source files (internal ellipses correctly marked); section attributions verified against the retrieved page structure (§III/§IV/§VI/§VIII offsets). - **Missed clauses: none that clearly prohibits.** The reviewer's independent sweep of all sources found no clause reaching the single-owner local-subprocess scope. - **Soundness: PASS.** The retraction of the first pass's function-based "proxy-style use" inference is justified by clause text, the §IV security chapeau, and the Terms' own deeming language ("deemed your actions", "extension of your own actions"). - **Classification: PASS is correct** under the owner's decision rule — no applicable clause prohibits the exact deployment; the two interpretive readings are supported by clause objects, not left genuinely ambiguous; exclusions retained. **Blockers found and fixed in `3be4b6b`** (both were update-discipline slips, not reasoning errors): 1. `docs/architecture.md` dependency paragraph still said "Stage 2 cancelled by D-047" in present tense → updated to "reopened by D-061". 2. `docs/providers.md` Z.ai-preset paragraph still asserted "the M07 NO-GO does not remove Z.ai HTTP/API support" in present tense → recast as "D-047's since-superseded M07 NO-GO never removed…". **Defers folded into the same remediation:** - §5.3 now acknowledges the S1/S2 wording asymmetry (S1 lacks the "unauthorized" qualifier) and resolves it by the clause chapeau plus the vendor's own automation surfaces. - §5.5 now name-checks the four additional swept clauses (traffic/engagement, competing-models, high-risk automated-decision, S2 Additional Terms for API Services) each with a not-reached rationale. - Evidence §7 exclusion list harmonized with D-061 point 6 (8 items, identical substance). - The S1 quote-verification extraction artifact was regenerated; every quoted fragment now verifies against the retrieval files. **Final verification on `3be4b6b`:** `uv sync --only-dev` OK; `make check` 2190 tests OK; `uv run basedpyright` 0 errors, 0 warnings, 0 notes; `git diff --check` clean. `origin/develop..HEAD` = exactly this issue's three commits (`112e950`, `8781060`, `3be4b6b`), no merge commit, all touched files in scope (docs only). UX impact: none — documentation only. Reviewer model: session model (GLM-5.3-Flash); not security-critical classification, so no additional security gate applies. PR remains open, unmerged, for human review.
adrian.tkacz deleted branch adapter/zcode-cli-stage2 2026-09-28 20:01:25 +00:00
Sign in to join this conversation.
Reviewers
No reviewers
adrian.tkacz
Labels
Clear labels   
accessibility
Barrier affecting people with disabilities

  
bug
Something isn't working

  
documentation
Improvements or additions to documentation

  
duplicate
This issue or pull request already exists

  
enhancement
New feature or request

  
good first issue
Good for newcomers

  
help wanted
Extra attention is needed

  
invalid
This doesn't seem right

  
P1
Owner-declared highest explicit implementation priority (loop ordering rank)

  
P2
Required product and integration work

  
P3
Deferred or producer-dependent work

  
question
Further information is requested

  
wontfix
This will not be worked on

No labels
accessibility
bug
documentation
duplicate
enhancement
good first issue
help wanted
invalid
P1
P2
P3
question
wontfix
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
2 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
Creatidy/scarcity-router!157
No description provided.