nexscope-cli 0.1.0__tar.gz

nexscope_cli-0.1.0/ACCEPTANCE.md

@@ -0,0 +1,525 @@
+# CLI acceptance record
+
+Date: 2026-10-26, Asia/Taipei. Current version: `0.1.0` (first independent version).
+The acceptance source baseline is `0.1.2.dev9` at
+`0252b0ec8a02ecdceb4deb691f3c3d0c12b8ed65`. The user selected `0.1.0` for the
+first independent version after dev11 source verification. Historical
+records below retain their original uncommitted/local-candidate context.
+
+This record covers implementation in the independent CLI repository, adapted
+from plugin commit `814f28016af6b3ff57ef12093a6bb133c4fb20b9`.
+It does not certify the entire R2 release or any published plugin package.
+
+## Implemented
+
+- Prefer documented numeric `extractedPrice`, falling back to numeric `price`
+  only when the first is absent/null; preserve zero, reject invalid numbers.
+- Preserve products with unqualified remote currency symbols. Comparable price
+  and currency remain unknown; `unresolvedCurrency`/`unresolvedPrice` and exact
+  response evidence preserve the source values. Never infer USD from `$`.
+- `workflow compare --resume` validates a completed candidates checkpoint,
+  imports its search evidence, rebuilds registered detail requests and applies
+  the original cumulative budget. Market, source type, auth mode and origin
+  remain bound. Historical search imports do not enter session caches.
+- Versioned datasets can generate HTML/CSV/Markdown offline while preserving
+  provenance, input hashes, errors and request accounting. Raw input receipts
+  and original dataset references have distinct meanings.
+- Bounded manifests, duplicate-key/non-finite checks, safe relative filenames,
+  link/reparse refusal, field validation, evidence/product/request/plan/result
+  consistency, non-overwriting outputs and portable report links.
+- Raw input limits: 10 MiB and 10,000 rows. Normalized datasets: 32 MiB;
+  complete evidence: 64 MiB. Large CSV fields respect the file limit and restore
+  the standard library's previous field limit afterward.
+- Public standard-library regressions, resource/version checker, wheel verifier,
+  six-job CI matrix, independent repository documentation and corrected notices.
+  Current resource hashes change; historical Skill source snapshot hashes do not.
+
+## Real-service acceptance and dev9 correction
+
+The user authorized real paid testing and separately confirmed the browser's
+connection action. An already logged-in **test-environment** account approved
+the connection; CLI credentials were stored only in Windows Credential Manager.
+The existing production environment key was not used. New persistent access was
+confirmed before creation, and the authorized test connection remains available
+for revocation in the test site's API Access page.
+
+Maximum authorized data calls: **4**. Actual data calls: **4** (one search, three
+details). CLI automatic retries: **0**. No further data request was issued after
+the budget was exhausted. Authentication/account reads are separate from these
+four data calls. No production database write, deployment or refund was made.
+
+| Invocation | Actual result | Confirmed charge |
+| --- | --- | --- |
+| dev8 keyword search, one page, `desk lamp` | Business 13007; upstream errcode 10000, no matching results; no candidates | 0 credits, provider diagnostic |
+| dev8 detail `B072MQ5BRX` | Server code 0 with one product, but CLI rejected `currency: "$"` as INVALID_PRODUCT | 90 credits, ledger |
+| dev8 detail `B08N5WRWNW` | Server code 0, empty products; CLI EMPTY_DATA | 10 credits, ledger |
+| dev9 detail `B072MQ5BRX`, one explicit verification call after the fix | COMPLETE / SUCCESS, exit 0, one product; no cache hit | 45 credits, ledger |
+
+**Total attributed charge: 145 credits**, confirmed by three test-ledger debit
+records and their user/route/time matches. The 10-credit empty-data debit remains
+a real charge; no refund was requested or performed. The whole account balance
+also changed because of concurrent activity, so its total delta was not used as
+the attribution measure. The bounded provider-log sample includes the original
+three attempts; the last verification is independently evidenced by its saved
+request/response and matching ledger time/route. Complete distributed-trace
+correlation and the registration/payment funnel are not certified here.
+
+The live defect was reproduced with a failing regression, then fixed in the
+shared validator and mapper. Remote currency metadata must be a string; an
+unqualified symbol is retained as unresolved metadata, while normalized currency
+and comparable price stay null. The saved real response revalidates offline with
+dev9 and preserves the original source price `15.27` and symbol `$`. The original
+dev8 FAILED record is unchanged, rather than relabeled as a successful run.
+
+Historical pre-publication wheel: `dist/nexscope_cli-0.1.2.dev9-py3-none-any.whl`, SHA-256
+`eaf560de2d949444b266308f9058acd679cfc622027e66f8ec616fb15f68a5e5`.
+Installed-wheel regressions: **30 pass on Windows and 30 on Ubuntu WSL2**;
+offline commands, loopback dual-auth, legacy adapters/runner, and dev8↔dev9
+upgrade/rollback pass. Windows's real isolated OS-store persistence and cleanup
+checks also pass. The live detail succeeded after these checks. Bundled adapter
+version pins and current hashes were updated; historical source snapshots were
+retained.
+
+Evidence is under `dist/live-20261026-1600/` (Git ignored): immutable dev8/dev9
+request bundles, filtered account snapshots, the four-call journal, read-only
+ledger evidence, bounded redacted log metadata, offline revalidation, and three
+original-failure audit report formats. `live-test-summary.json` indexes hashes
+and results. Secrets and raw unfiltered logs are excluded. The failure reports
+faithfully retain zero successful products from the original run; they are not
+presented as a successful comparison.
+
+**Disposition: partial acceptance.** Real account connection, valid live detail,
+charge reconciliation and the currency correction pass. Successful keyword
+selection/resume and a complete live multi-product comparison remain unaccepted
+because the search returned no candidates and the original second detail was
+empty. Mock/replay checks do not substitute for those missing live outcomes.
+
+## Initial local verification (before the additional acceptance below)
+
+- All 28 regression tests pass on Windows with Python 3.11.9 and 3.12.14.
+- Actual wheel built with Python 3.12.14, pip 26.2.1, setuptools 84.0.0.
+- Fresh venv outside the checkout; actual wheel installed without editable mode
+  or dependencies. Source path injection removed; imports checked against
+  `site-packages` in that environment. Wheel subpackages, resources and license
+  files inspected. No bytecode or plugin metadata included in the wheel.
+- Network and keyring/requests imports blocked for offline checks: template,
+  samples, schema, doctor, plans, three report formats, versioned re-exports,
+  fictional three-ASIN task, keyword selection/resume and partial failure.
+  Paths contain spaces and Chinese characters.
+- 10,000-row input and a near-10-MiB input with a larger normalized output;
+  rejection of row/byte overflow, malformed input and inconsistent evidence.
+- Fictional loopback HTTP service exercises both auth contexts, keyword resume,
+  price-only detail responses and authenticated redirect refusal. Browser auth
+  uses an in-memory fake keyring backend. It never accesses a real OS credential
+  store, service account or billable Nexscope endpoint.
+- Standalone installed legacy scripts preserve page/batch input, flags and env
+  behavior. The legacy browser `run` path also passes with reviewed hash-pinned
+  dependencies installed from PyPI. No arbitrary dependency is auto-installed
+  during a query.
+- Explicit old wheel supplied to the verifier: install `dev7`, upgrade to `dev8`,
+  roll back to `dev7`, restore `dev8`; command version checks pass. User-report
+  and test environment-key sentinels survive. This does not test real OS-keyring
+  credential persistence across installer operations.
+
+## Public source and hosted CI
+
+The repository is public at <https://github.com/nexscope-ai/nexscope-cli> and the
+initial implementation was pushed to `main` as commit
+`d5ed885ef87a248f82827a8c6252760e340d7a7a`. Its
+[installed-wheel workflow](https://github.com/nexscope-ai/nexscope-cli/actions/runs/40296830306)
+completed successfully on Windows, Ubuntu and macOS with Python 3.11 and 3.12:
+all six jobs passed. These are builds from an immutable Git commit. Private
+live-test artifacts and credentials were excluded from the commit and push.
+
+This closes the corresponding hosted installation/offline/loopback matrix,
+including Windows Python 3.11. It does not certify a local Mac's Keychain,
+real-account connection, real comparison workflows, Linux desktop credential
+services, refunds or host integrations. No package release, tag or plugin upload
+was made. See `docs/HANDOFF-MACOS.md` for the remaining Mac work.
+
+## Additional executable acceptance, 2026-10-26
+
+The same wheel was tested again, without changing its runtime code:
+`nexscope_cli-0.1.2.dev8-py3-none-any.whl`, SHA-256
+`653e86eb614ee649fddd58b10b36631dc9ce2c949af534c6b3371fc9b18b1024`.
+
+| Acceptance item | Evidence and outcome |
+| --- | --- |
+| Windows external wheel installation | PASS: Python 3.12.14; 28 installed regressions, offline CLI/report/workflow cases, loopback dual-auth and legacy scripts/runner |
+| Windows real Credential Manager | PASS: actual `keyring.backends.Windows`, unique fictional service entry, write/read and fingerprint verification; original account entries untouched |
+| Windows credential persistence | PASS: test credential survives dev8→dev7→dev8 and package uninstall/reinstall; final deletion confirmed |
+| Ubuntu WSL2 external installation | PASS: Python 3.12.3, glibc 2.39, Linux 6.6.87.2 WSL2; 28 installed regressions, offline cases, loopback dual-auth and legacy scripts/runner |
+| Ubuntu WSL2 upgrade/rollback | PASS: dev7→dev8→dev7→dev8, executable version checks and user-report preservation |
+| Linux desktop credential store prerequisite | BLOCKED: actual backend is `keyring.backends.fail`; CLI rejection confirmed (`UNAVAILABLE_FAIL_CLOSED`); no plaintext fallback |
+| Real test API unauthenticated account request | PASS for negative authentication only: GET `/api/cli/account`, HTTP 200, business code 12003, no data returned; one request, no credentials sent |
+| Live-data prerequisites before paid-test approval | The configured environment key belonged to production and was not used; later paid testing used a separately approved test browser connection as recorded above |
+
+Ubuntu's system Python lacks `ensurepip`, and passwordless sudo was unavailable.
+The verifier instead created an ordinary isolated venv and bootstrapped pip from
+the explicit Python-bundled `pip-25.0.1-py3-none-any.whl`, SHA-256
+`c46efd13b6aa8279f33f2864459c8ce587ea6a1a59ee20de055868d8f7688f7f`.
+Application imports still came exclusively from the new venv's `site-packages`.
+No system package, sudo configuration or WSL configuration was changed.
+
+Repeat the Windows OS-store check with:
+
+```sh
+python scripts/verify_wheel.py dist/nexscope_cli-0.1.2.dev8-py3-none-any.whl --previous-wheel /path/to/nexscope_cli-0.1.2.dev7-py3-none-any.whl --with-dependencies --with-os-keyring
+```
+
+For WSL2, run the same verifier with its Linux Python, use mounted absolute wheel
+paths and supply `--pip-wheel /path/to/reviewed/pip.whl` if ensurepip is missing.
+`--with-os-keyring` requires a working real desktop credential store; the regular
+dependency check also records availability and verifies fail-closed rejection.
+
+Evidence files, ignored by Git and containing no credential values:
+
+- `dist/verification-windows-os-keyring-py312.txt`
+- `dist/verification-linux-wsl2-py312.txt`
+- `dist/live-route-prerequisites.json`
+- `dist/credential-prerequisites.json`
+- `dist/acceptance-20261026.json` (artifact/test/verifier fingerprints)
+
+That pre-live acceptance changed only verification scripts and documentation.
+Its dev8 artifact and provenance remain unchanged. The later live correction is
+separately versioned dev9, with its own wheel and checks recorded above.
+
+The historical local wheel hashes and external-install transcripts are saved in
+`dist/`, which is ignored by Git. Those tests predated source publication. Record
+`git rev-parse HEAD` and the actual built wheel hash for each subsequent run;
+documentation and build timestamps can change wheel bytes. A versioned release
+still requires review and verification from an immutable source commit.
+
+## Outstanding release gates
+
+| Gate | Status |
+| --- | --- |
+| Windows fresh wheel, offline and loopback checks | Locally verified as described above |
+| Real Windows OS keyring | Verified with isolated synthetic credentials, including upgrade/rollback/uninstall and cleanup |
+| Real Nexscope login and valid test-account calls | PASS for an existing logged-in test account, approved CLI connection, and one successful dev9 live detail; new-user registration is not tested |
+| Linux WSL2 install, offline, loopback and upgrade/rollback | Verified locally on Ubuntu Python 3.12.3; does not certify all Linux hosts |
+| Linux Secret Service/KWallet | BLOCKED on this headless WSL2 host: `keyring.backends.fail`, fail-closed rejection verified; no desktop-store persistence acceptance |
+| macOS install, OS keyring and upgrade/rollback | PASS locally for arm64/macOS 27.0, Python 3.11.15/3.12.13; real Keychain dev9↔dev10 and uninstall persistence/cleanup on Python 3.12.13, with an explicitly approved browser connection; see completion record below |
+| Real search/detail response shapes, balance, pricing, billed/refund outcomes | PARTIAL: Mac live comparison and 2-success/1-failure case pass with 6 calls / 230-credit reconciliation; search failed, so live resume is BLOCKED. Earlier Windows 145 credits are separate. Refund/server-attribution gates remain unaccepted |
+| Codex/Claude Code natural-language three-ASIN, keyword and partial tasks | Offline/mock tasks pass in both actual hosts. Claude dev10 and fresh configured Codex dev11 live direct/partial pass. Live search/resume and published-plugin gates remain unaccepted; see fresh-session correction below |
+| Registration → effective call → payment/refund server attribution | Existing CLI metadata only; backend/web chain not implemented or certified here |
+| Named pilot owners, compatibility window, support and acceptance sample | Needs product/release assignment |
+| Source commit and CI | Verify the checked-out commit with `git rev-parse HEAD` and inspect its actual CI results; the local evidence above predates source publication |
+| Versioned releases and downstream Skills/plugin migration | No package release, tag or plugin upload; separate acceptance and release authorization remain required |
+| Official published plugin contents | Not inspected or modified |
+
+Do not mark the R2 release accepted, replace existing published Skills, or
+describe fictional fixtures as live acceptance until the relevant gates pass.
+The three known CLI defects are addressed; these external gates remain explicit.
+
+## Local macOS free-phase record, 2026-10-26 (Asia/Taipei)
+
+Actual checkout: `main`, HEAD `0252b0ec8a02ecdceb4deb691f3c3d0c12b8ed65`;
+new clone, initially clean. No pre-existing checkout or user changes were replaced.
+Platform: **macOS 27.0 (26A428), arm64**. Executed interpreters:
+Python **3.11.15** (uv) and **3.12.13** (mise). System Python 3.9.6 was not used
+for the package. No x86_64, other local OS, or post-change hosted CI claim is made.
+
+The final dev10 candidate is that HEAD **plus local uncommitted fixes and version
+pins**, not an immutable dev10 Git commit. No commit, push, tag, release, PyPI
+publication or downstream plugin update was performed. Wheel bytes were compared
+against every current packaged source/resource file and matched.
+
+| Artifact | Actual SHA-256 |
+| --- | --- |
+| Initial Mac dev9 wheel, `dist/nexscope_cli-0.1.2.dev9-py3-none-any.whl` | `55606905a9c5acf5c0d5c7e63c517a7525109e825922a0388dc8446ff73ea13c` |
+| Rebuilt, verified dev9 predecessor, `dist/rebuilt/nexscope_cli-0.1.2.dev9-py3-none-any.whl` | `618c18c6c7e134eb992479242d60fa672d9ac99bfd2519b8b3529bc2d7fff02e` |
+| Final local dev10, `dist/dev10/nexscope_cli-0.1.2.dev10-py3-none-any.whl` | `eb36658d08cf523c714444bb82ce5fc3acf16a305248fbcbd688f3c5fc25744e` |
+
+### PASS
+
+- `python scripts/check_resources.py`: resource integrity and paired adapter
+  pins pass. Historical `sourceSnapshot` hashes are unchanged.
+- Actual wheels installed into fresh environments outside the checkout;
+  source-path injection removed and imports checked against `site-packages`.
+  Dev9: **30 installed regressions per interpreter**. Dev10: **31 per
+  interpreter**, offline reports, space/non-ASCII paths, failure exits, output
+  preservation and fictional loopback dual-auth checks pass.
+- Real `keyring.backends.macOS` on Python 3.12.13: unique synthetic credential
+  write/read, dev9→dev10→dev9→dev10 persistence, uninstall/reinstall retention,
+  and deletion verification pass. Only verifier-owned synthetic entries were
+  deleted; Keychain protections were not changed. Python 3.11 rollback/version
+  and user-report retention pass; real-store persistence was exercised on 3.12.
+- User logged into the test frontend and explicitly approved **Connect
+  Nexscope** before a persistent CLI Key was created. Installed dev10 completed
+  `auth poll`; browser and CLI account matched. Credential exists in Keychain,
+  pending credential was removed, and its value was absent from this run's
+  controlled text artifacts. This is not a whole-disk secret audit.
+- Browser-account `account` with an unrelated environment sentinel still
+  matched the browser account. Installed-wheel negative checks show that an
+  empty browser namespace does not use an environment key, and missing env-mode
+  variables neither read Keychain nor fall back to the connected browser.
+  Both negatives exited 1 with network blocked. Loopback positive dual-auth
+  coverage remains distinct from two real-account positive modes.
+- Actual **Codex CLI 0.163.0** and **Claude Code 2.1.308** received natural-language
+  tasks and drove the installed dev10 CLI: three-product COMPLETE (3 successes,
+  exit 0, 3 mock requests), keyword CANDIDATES_READY (1), saved `--resume`
+  COMPLETE (2 new requests, cumulative 3, exit 0), PARTIAL (2 successes + 1
+  EMPTY_DATA, exit 1, 3 requests), all three report formats and offline
+  re-export. Search evidence was byte-identical on resume, and ASIN-selected
+  mock detail fields now match their candidates. These are **offline/mock host
+  runs**, not live data or published-plugin acceptance. Both hosts also ran the
+  original dev9; its original outputs and the discovered mismatch are retained.
+- Own-account **API Access → Call Details** provides a read-only reconciliation
+  entry with UTC time, masked key, route, actual credits, request and diagnostic
+  IDs. Only older calls appeared in the observed latest page; the new Mac key
+  had no data-call row. No CSV/key export, credential discovery or database access
+  was used. The whole account balance delta was not attributed to this task.
+
+### FAIL, preserved and corrected
+
+- Original `verify_wheel.py` copied the Python executable. Both local Python
+  distributions aborted with SIGABRT; direct reproduction showed dyld could not
+  find relative `libpython3.12.dylib`. A failing verifier regression was added
+  before the fix. Using the platform's normal venv symlink behavior on POSIX
+  fixed both interpreters; Windows retains copies. The external-wheel regression
+  and final installed checks pass. Original failure logs remain unchanged.
+- Actual Claude host inspection exposed known mock ASINs receiving another
+  fixture's title/price in detail responses. The new regression failed for the
+  three candidates before the correction. Dev10 selects a matching fixture by
+  ASIN, preserving the existing deterministic fallback for arbitrary fictional
+  IDs. Both hosts re-executed against the rebuilt dev10 and confirmed matching
+  fields. Static mock keyword data remains intentionally fictional.
+
+### BLOCKED / separate gates
+
+- Live successful multi-product comparison, keyword/resume, real partial
+  completion and corresponding live-host tasks: **this free-phase checkpoint preceded the later 20-call authorization;
+  see the post-approval record below for actual outcomes**. Prior four-call authorization is
+  exhausted. No new live product/search request has been sent.
+- A second real env-api-key account has not been provisioned or used. Positive
+  real env-mode account isolation is unaccepted; no browser credential was
+  exported into the environment to simulate it.
+- Refunds, complete server-side attribution, registration/payment funnel,
+  Linux desktop keyrings, named pilot/support ownership and downstream release
+  gates remain separate. Existing Codex host startup reported a stale
+  `nexscope_test` MCP OAuth `invalid_grant`; offline CLI tasks used no MCP query
+  and no host settings were changed. Published-plugin discovery was not tested.
+
+New **Nexscope data calls: 0; automatic data retries: 0**. Five CLI auth/account
+HTTP requests were made: one start, one poll, one acknowledgement and two account
+reads; browser login/approval traffic is separate. No newly attributed data
+debit exists in this run's evidence; this is not a complete account-ledger audit.
+The old **145 credits** remain historical and are not charged to this Mac run.
+Claude's two host runs separately reported model list-price costs of
+USD 0.4659312 and 0.6272848; these are not independently reconciled charges or
+Nexscope credits. Codex supplied token usage without a monetary amount.
+
+Commands (replace `python` with the recorded 3.11 or 3.12 interpreter):
+
+```sh
+python -m pip wheel --no-deps --wheel-dir dist/dev10 .
+python scripts/check_resources.py
+python scripts/verify_wheel.py dist/dev10/nexscope_cli-0.1.2.dev10-py3-none-any.whl --previous-wheel dist/rebuilt/nexscope_cli-0.1.2.dev9-py3-none-any.whl --with-dependencies --with-os-keyring
+python scripts/test_verify_wheel.py dist/dev10/nexscope_cli-0.1.2.dev10-py3-none-any.whl
+```
+
+Python 3.11 ran the verifier with `--previous-wheel`, without the two keyring/
+dependency flags. Private evidence: **`dist/macos-20261026/` (Git ignored)**,
+including original failures, build/install/verification transcripts, auth/account
+snapshots without keys, read-only billing table, host prompts/events, copied host
+outputs, `acceptance-summary.json` and file hashes. Offline reruns are explicitly
+labeled and never replace original failures. The external `live-env` and host
+working directories under `~/.local/share/nexscope-cli-acceptance/20261026/` are
+retained for pending acceptance; no task worktrees or branches were created.
+
+## Post-approval live acceptance on this Mac
+
+The user explicitly expanded authorization to **20 data calls, no credit ceiling**,
+and delegated sample selection. Fixed samples: `B072MQ5BRX`, `B07FZ8S74R`, with
+`B08N5WRWNW` solely as the historical failure sample; one keyword,
+**`wireless mouse`**, on `amazon.com`. All runs used the installed dev10 wheel
+SHA-256 `eb36658d08cf523c714444bb82ce5fc3acf16a305248fbcbd688f3c5fc25744e`
+outside the checkout, origin `https://test-claw-callback.nexscope.ai`, and the
+approved Mac browser-account Keychain connection. Source remains the recorded
+dev9 HEAD plus local uncommitted dev10 changes. No production call occurred.
+
+| Gate | Status and evidence |
+| --- | --- |
+| Actual Claude Code direct two-product workflow | **PASS**: COMPLETE, 2/2 successes, exit 0; 2 requests, no cache hits/retries; three report formats and original responses |
+| Single live keyword search | **FAIL, upstream**: business 13007, empty candidates, exit 1; 1 request; original failure preserved |
+| Live keyword `--resume` | **BLOCKED**: no completed CANDIDATES_READY checkpoint; no duplicate search, altered keyword or extra selection |
+| Actual Claude Code partial workflow | **PASS**: B072MQ5BRX and B07FZ8S74R succeed; B08N5WRWNW EMPTY_DATA; PARTIAL, 2/3 successes, exit 1, 3 requests; three formats retain errors |
+| Live dataset offline re-export | **PASS**: direct and partial outputs re-export into HTML/CSV/Markdown, preserving source data, original request bytes/hashes, errors, accounting and relative evidence links; clearly offline revalidation |
+| Actual Codex CLI live execution | **BLOCKED**: workspace-write host sandbox produced KeyringError before workflow artifacts; the host execution failed and was not retried. No Keychain or sandbox protections were weakened. Offline host checks remain separate PASS |
+| Bounded read-only billing reconciliation | **PASS**: six new-key call-detail rows and new-key summary match 6 requests / 230 charged credits; separate from old key's 145 credits |
+
+**Actual data calls: 6 / 20; unused: 14; automatic retries: 0.** Five details and
+one search were sent. The direct result contains coffee and Echo Dot products;
+this is functional comparison acceptance, not a claim they are competing
+products. Both remote `$` currencies remain unresolved, prices unknown for
+comparison, with original numeric prices retained as evidence.
+
+| Phase, UTC 2026-10-26 | ASIN/query | Actual CLI outcome | Visible charged credits |
+| --- | --- | --- | --- |
+| Direct, 10:02:03 | B072MQ5BRX | SUCCESS | 45 |
+| Direct, 10:02:04 | B07FZ8S74R | SUCCESS | 90 |
+| Search, 10:05:00 | wireless mouse | BUSINESS_13007, no candidates | `—` in call detail; no additional numeric debit in matched key summary |
+| Partial, 10:06:08 | B072MQ5BRX | SUCCESS | 45 |
+| Partial, 10:06:09 | B07FZ8S74R | SUCCESS | 45 |
+| Partial, 10:06:10 | B08N5WRWNW | EMPTY_DATA | 5 |
+
+The five numeric detail charges total **230 credits**, matching the new key's
+six-call summary. The empty-data debit is real and was not refunded. Do not
+translate the search detail's dash into an independently certified zero, or use
+response `cost` as credits. API usage labels all six admitted requests SUCCESS;
+that label does not override the saved CLI business failure or empty result.
+Matching uses the isolated new key, UTC intervals and routes; private billing
+request/diagnostic IDs are retained. Detail payload trace IDs were null, and
+the search provider trace ID is not asserted equal to the billing diagnostic ID.
+Full distributed tracing and refund/server attribution remain separate gates.
+
+Earlier account call-table snapshots had no new rows. A fresh page and current
+Last 7 days query subsequently exposed all six; those initial snapshots remain
+unaltered, and their temporary billing BLOCKED status is superseded by this
+bounded reconciliation. The account balance also changed with concurrent work;
+its delta was never used as this run's expense.
+
+Installed commands were driven by natural-language prompts in actual Claude
+Code, with the CLI pinned and no MCP data calls. Each phase saved a dry-run and
+one execution: direct `--asins B072MQ5BRX,B07FZ8S74R --max-calls 2`; search
+`--keyword "wireless mouse" --max-calls 3`; partial
+`--asins B072MQ5BRX,B07FZ8S74R,B08N5WRWNW --max-calls 3`. Executions used
+`--market amazon.com --auth-mode browser-account --execute --allow-paid --format all`.
+Exact commands, UTC times and exits are in the private transcripts.
+
+Private evidence: `dist/macos-20261026/live/`, including original host failure,
+live bundles, offline re-exports, call-table/key-summary text, `live-summary.json`
+and hashes. External `live-codex/` and `live-claude/` directories under the
+recorded acceptance root are retained for pending gates. The persistent approved
+test connection remains in Keychain and can be revoked in API Access. No key
+value was exported into the environment or a file, no Windows credentials were
+copied, and no refund/database write/deployment/publication was performed.
+
+**Disposition remains partial acceptance.** Do not spend the remaining allowance
+on automatic retries, new keywords or extra ASINs. Successful live search/resume,
+supported Codex live credential access, positive real env-mode isolation, refund/
+full server attribution and downstream release gates remain unaccepted.
+
+### Final local dev11 correction and offline verification
+
+Claude's partial-case offline re-export found another preflight defect:
+`report compare --dry-run` returned PLAN_READY for a missing output parent,
+while export failed. The existing output-preservation regression was extended
+and failed before the fix. Dev11 checks the parent directory and output/evidence
+collisions before either dry-run or export, without creating directories or
+overwriting files. Repeated export after explicitly creating a parent is labeled
+an **offline rerun**, never a new live call. Original host/export failures remain.
+
+Final wheel: **`dist/dev11/nexscope_cli-0.1.2.dev11-py3-none-any.whl`**, SHA-256
+**`5c94b388da7f691841aef346195806887fae9f1a6148f62aed130f6c1f2afa5c`**.
+Its packaged bytes match the current local source/resources. Source HEAD is
+unchanged; this remains an uncommitted local candidate. Version pins and current
+adapter hashes match dev11; historical snapshots are unchanged.
+
+**PASS**: source 31 regressions; actual dev11 installed wheel 31 regressions on
+each Python 3.11.15/3.12.13; offline/loopback cases and resource checks; 3.12
+hash-pinned dependency/legacy checks plus real macOS Keychain dev10↔dev11
+write/read/rollback/uninstall retention/cleanup; 3.11 version/report rollback.
+The separate external-environment verifier regression also passes. Commands:
+
+```sh
+python scripts/check_resources.py
+python scripts/verify_wheel.py dist/dev11/nexscope_cli-0.1.2.dev11-py3-none-any.whl --previous-wheel dist/dev10/nexscope_cli-0.1.2.dev10-py3-none-any.whl --with-dependencies --with-os-keyring
+python scripts/test_verify_wheel.py dist/dev11/nexscope_cli-0.1.2.dev11-py3-none-any.whl
+```
+
+The 3.11 verifier again omitted dependency/OS-keyring flags. Installed dev11
+re-exported both original live dev10 datasets into all three formats offline,
+preserving products/errors/accounting and every original request byte. Missing
+parent dry-runs now exit 1 without creating outputs. **No dev11 live request was
+sent**; live host outcomes above belong to the exact dev10 wheel, not dev11.
+`live/dev11-offline-revalidation.json` records this distinction.
+
+Additional Claude live-stage model usage reported USD 0.7058736 (direct),
+0.2560884 (keyword) and 0.695398 (partial). These provider-reported list-price
+figures are not independently reconciled payments and are separate from the
+230 Nexscope credits. All raw live/host/export failures, final transcripts,
+source diff and SHA-256 index remain in the ignored evidence directory.
+
+## Fresh Codex session retest and permission-profile correction
+
+The user requested a new session after suspecting a changed conversation key,
+then explicitly narrowed further work to **Codex only**. No old Codex session
+was resumed. The original `KeyringError` records remain unchanged.
+
+**PASS / test-setup correction:** earlier child Codex runs had been launched
+with an extra `--sandbox workspace-write` override. The user's existing local
+configuration was already `sandbox_mode = "danger-full-access"`. A local
+diagnostic reproduced `keyring.backends.macOS` → `KeyringError` → native
+`Error(-50)` under the extra sandbox; Codex's denial capture reported blocked
+`mach-lookup com.apple.securityd.xpc` and `com.apple.SecurityServer`. The same
+credential read succeeded in the current host. This identifies an added test
+restriction rather than an expired Nexscope connection. An initial diagnostic
+script mistakenly called `client.store.load` and raised AttributeError; that
+diagnostic failure was retained, corrected to `client.load`, and is not treated
+as a CLI defect.
+
+A fresh `codex exec --ephemeral --skip-git-repo-check --json` run, **without a
+sandbox override**, inherited the user's existing permissions. No global config,
+Keychain protection or OS security setting was changed; no new key was created,
+and no credential was exported. Installed CLI was dev11, with the same recorded
+wheel hash and unchanged local source HEAD.
+
+| Fresh configured Codex gate | Outcome |
+| --- | --- |
+| Normal `account` | **PASS**, existing approved Mac account connected |
+| Direct B072MQ5BRX/B07FZ8S74R | **PASS**, COMPLETE, 2 successes, exit 0, 2 requests |
+| Single `wireless mouse` search | **FAIL**, BUSINESS_13007, exit 1, 1 request, no candidates |
+| Real keyword `--resume` | **BLOCKED**, no completed candidates; not a Codex conversation-resume operation |
+| Fixed three-ASIN partial case | **PASS**, 2 successes + B08N5WRWNW EMPTY_DATA, PARTIAL, exit 1, 3 requests |
+| All reports and offline re-export | **PASS**, direct/partial HTML/CSV/Markdown; manifests and original request bytes preserved |
+
+This closes **Codex live account/direct/partial acceptance for the user's current
+configured profile**, not Keychain access under workspace-write restrictions.
+New conversation creation does not resolve the server search error. A bounded
+read-only log query for the new search trace returned zero matches; its deeper
+provider cause remains unconfirmed. No backend fix or retry was attempted.
+
+Before the user narrowed scope, one extra fresh Claude session re-tested only
+the failed keyword (not its already-passing direct/partial cases): 1 request,
+BUSINESS_13007, no resume. This extra call is explicitly counted. No further
+Claude task was launched after the Codex-only instruction.
+
+**New configured Codex requests: 6, charged 185 credits** (direct 45+45;
+partial 45+45+5). The additional search detail rows display `—`; new-key summary
+reconciles **13 cumulative Mac calls / 415 charged credits**. This is the earlier
+6/230, one extra failed Claude search, plus Codex 6/185. **7 of 20 calls remain
+unused; automatic retries 0.** Each repeated search was an explicit user-requested
+fresh-session retest, not an automatic retry. The historical Windows key's 145
+credits remain separate; account balance deltas and response `cost` are unused.
+
+Private original/fresh-session prompts, exits, diagnostic exceptions/denials,
+live dev11 bundles, offline reports and read-only billing rows are under
+`dist/macos-20261026/retest/`, indexed by `retest-summary.json` and the parent
+SHA-256 index. Original failures are retained. External `retest-codex/`,
+`retest-codex-configured/`, and `retest-claude/` directories are retained for
+reproducibility. No source/runtime change or publication was needed for this
+retest. Remaining gates include live successful search/resume, positive real
+env-key isolation, full backend/refund attribution and release authorization.
+
+## First independent version: 0.1.0
+
+At the user's request, the current version is **0.1.0**, replacing the development
+version label. Package metadata, module version, dependency locks, paired Skill
+requirements/adapter pins, installation instructions and CI wheel path agree.
+Historical test versions and original failures above are preserved as recorded.
+No runtime behavior or outstanding acceptance status changed with this renumbering.
+
+Built from source baseline `430461255eff13a2119486912682742e68d7582e` plus this
+version change: `dist/version-0.1.0/nexscope_cli-0.1.0-py3-none-any.whl`, SHA-256
+`5703a4c748d32759903270fb4fbeeb2b84ea5e905e3c7dd29204eb551c8a9a47`.
+Resource checks pass. Actual external installed-wheel verification passes on
+Python 3.11.15 and 3.12.13: **31 regressions each**, offline/loopback checks and
+version-transition/report preservation against the exact prior dev11 wheel.
+Python 3.12 additionally passes hash-pinned dependencies/legacy checks and real
+Keychain write/read, version-transition/uninstall persistence and cleanup.
+This is a version transition, not a claim that 0.1.0 sorts above 0.1.2.dev11.
+No new paid call was made; cumulative usage remains 13/20 and 415 credits.
+Private transcripts are in ignored `dist/version-0.1.0/`. Source push is
+authorized; PyPI, GitHub Release assets/tags and plugin publication are separate.

nexscope_cli-0.1.0/ASSET-NOTICES.md

@@ -0,0 +1,9 @@
+# Bundled asset attribution
+
+Copyright 2026 Nexscope.
+
+This independent CLI contains report templates and fictional product samples.
+Templates use system fonts, with no external font, JavaScript or image assets.
+The original plugin icon and plugin entrypoint are not part of this distribution.
+Apache-2.0 Section 6 governs trademark use; the copyright license does not grant
+trademark rights or imply Nexscope endorsement of modified distributions.