@n8r/surface 0.0.0-stage → 0.1.0-experimental.5

package/docs/n8r-surface/researcher-skill.md

@@ -0,0 +1,309 @@
+# n8r-surface Researcher: start an investigation
+
+For candidate 0.1.14, use the [first-use overview check](first-use-overview.md).
+The [0.1.13 tester guide](tester-candidate.md) retains pinned baseline installation
+and host boundaries; use the exact 0.1.14 PR revision for the new check. Installation versions and
+branch references below describe historical rehearsals, not the current candidate.
+
+**Previously accepted standalone skill 0.1.11:** produces working notes and checked
+Capability/Bound Surface bundles using the host's source access and conversation.
+The skill and [accepted two-product verification](skill-usefulness-verification.md)
+are merged in PRs #54 and #58. It needs neither the `n8r-surface` executable nor
+an npm release. Related to [#17](https://github.com/injectionator/n8r-surface/issues/17).
+
+The team installation below pins merged commit
+[`5eed2ac9a31933caa2ae306e4b407e327d147bf7`](https://github.com/injectionator/n8r-surface/commit/5eed2ac9a31933caa2ae306e4b407e327d147bf7).
+Its plugin files are identical to the accepted trial revision `8e4cb3a`.
+This is an identified repository distribution; no separate skill 0.1.11 tag or
+GitHub Release has been published. Tagging, uploading and publication require
+separate authorization. Skill delivery remains first; CLI and Labs work is deferred.
+
+**Verified host:** Codex CLI 0.166.0 app-server with ChatGPT sign-in and
+`gpt-6-astra`; the accepted artifact trial used Node.js 26.13.0. Checked artifacts
+and workspace helpers require Node.js 22, 24 or 26 and persistent filesystem
+access. ChatGPT desktop/Work picker, invocation and uninstall/reinstall acceptance
+remain pending. A CLI installation result does not establish desktop availability.
+
+The primary result is a partial **AI Surface graph**, with evidence and unresolved
+connections. It is not an inventory of every application feature or setting.
+Configuration matters when it establishes or explains a relevant node, edge,
+route, data flow or authority boundary. The bundled
+[discovery guide](../../plugins/n8r-surface/skills/research/references/discovery-map.md)
+covers available source combinations, reusable product knowledge and comparison.
+See [discovery, comparison and live-inspection examples](researcher-discovery.md)
+for those broader tasks; the short first trial below uses authentication.
+For a compact security brief and explicit reconciliation, use the
+[Skeleton Surface walkthrough](researcher-skeleton-check.md). This provisional
+Markdown profile adds no validated structured-output contract.
+
+For workspace-only setup and recognition, use the
+[workspace contract](workspace.md), [isolated candidate installation](workspace-installation.md)
+and [short human check](workspace-human-check.md). The workspace helper requires
+Node.js 22, 24 or 26 and a persistent local filesystem. Desktop workspace
+acceptance is pending; a cloud-only chat cannot perform local setup.
+
+## Install or update 0.1.11
+
+Prerequisites: Git, Codex CLI 0.166.0 and network access to GitHub. These commands
+change the selected Codex home's plugin registration and cache. Research uses your
+existing ChatGPT sign-in; installation itself needs no inference credentials.
+No manual clone, npm install or build of n8r-surface is needed.
+
+For an existing installation, first record its source and version:
+
+```sh
+codex plugin list --marketplace n8r-surface --json
+```
+
+When replacing an earlier Git or local marketplace registration, remove only this
+marketplace registration before running the pinned installation below:
+
+```sh
+codex plugin marketplace remove n8r-surface --json
+```
+
+This removes the marketplace snapshot, not research saved outside the plugin cache.
+Skip removal for a first installation. A commit-pinned marketplace stays pinned;
+`marketplace upgrade` does not select a newer commit. Use this remove/add procedure
+with the next approved revision when deliberately updating. Keep the old source/ref
+if you will need to restore it. Refreshing the app alone does not update the snapshot.
+
+From any directory:
+
+```sh
+codex plugin marketplace add injectionator/n8r-surface --ref 5eed2ac9a31933caa2ae306e4b407e327d147bf7 --json
+codex plugin add n8r-surface@n8r-surface --json
+codex plugin list --marketplace n8r-surface --json
+```
+
+Expected: each command exits 0; the list has exactly one installed/enabled
+`n8r-surface@n8r-surface`, version `0.1.11`. The installation path varies.
+For a rehearsal that leaves your active installation alone, use the temporary
+`CODEX_HOME` procedure in [isolated installation](workspace-installation.md#isolated-maintainer-installation),
+substituting the pinned revision and expected version above. That page's 0.1.9
+results are historical.
+
+## RESEARCH-01: investigate and continue
+
+1. **Check the installation.** Complete the installation above and verify 0.1.11.
+   Use a local Codex chat with access to the selected checkout and writable notes
+   folder. Work-mode trials remain subject to the separate
+   [human acceptance](researcher-skill-verification.md#remaining-work-mode-acceptance).
+2. **Open your source.** Start a local chat with your LibreChat folder selected.
+   In Work, select **n8r-surface Researcher** via `@`; the qualified skill is
+   `n8r-surface:research`. Refresh/restart the app if the installed version is
+   missing. The host must actually read the checkout; naming a path in a cloud
+   chat does not supply its files.
+3. **Ask one bounded question:**
+
+   > Use n8r-surface Researcher to investigate authentication in this LibreChat
+   > checkout. Explain the components and configuration choices, cite the files
+   > you read, and distinguish supported options from my selected setup. Save
+   > working notes in your default research folder, show me where they are,
+   > and ask the next useful question.
+
+   Expect a source revision, an explanation with file/line evidence, unanswered
+   questions and a saved Markdown note. Supported options are not your selected
+   deployment, and reading source does not verify a running integration.
+4. **Open the saved note.** The default on a local filesystem is
+   `~/Documents/n8r-surface/research/librechat/authentication-001.md`.
+   The Researcher shows the resolved absolute destination before saving and
+   provides an openable link after reading it back. Existing files get the next
+   unused number. You can request another folder. A host without persistent local
+   file access must say so and provide a download or text; that does not pass the
+   local-save check.
+5. **Answer, then continue.** Answer the Researcher's actual question or correct
+   one finding. Ask it to save a successor. Expect `authentication-002.md` (or
+   the next unused number), the dated human decision separated from source
+   evidence, and the first note unchanged. In a fresh chat with the same source
+   access and skill selected, provide that successor and ask:
+
+   > Continue this research from the saved note. Check the source revision,
+   > retain my decisions and unresolved questions, and investigate the next
+   > useful gap. Save a successor and show me the result.
+
+Wording, source revision, findings, date and file number may vary. Missing access,
+invented citations, a missing saved file, overwritten earlier notes or a lost human
+decision are failures. Independently opening the files is the readback check;
+an agent rehearsal is separate from human acceptance. No service is started and
+the source checkout should remain unchanged. Keep the notes after the trial;
+plugin removal does not delete them.
+
+**State X:** installed skill, readable identified checkout, and a writable notes
+folder. **Input Y:** the prompt above, followed by an actual human answer and a
+fresh-chat continuation. **Outcome Z:** a useful explanation and preserved,
+readable successor notes with source evidence, choices and gaps. See
+[observed results and remaining limits](researcher-skill-verification.md).
+
+These checks cover working notes. Skill 0.1.11 also provides the separate
+[checked artifact workflow](skill-surface-artifacts.md); Lab adaptation remains
+outside this repository. RESEARCH-01 does not establish the
+complete installed-command E2E-01/E2E-02 cases.
+
+## Package and standards
+
+The canonical source is [the plugin folder](../../plugins/n8r-surface/).
+It contains one `skills/research/SKILL.md`, bundled references, optional
+OpenAI skill metadata, licence files and an icon. It has no MCP server, hooks,
+installer scripts, embedded credentials or separate model connection.
+
+The [Agent Skills specification](https://agentskills.io/specification) defines
+`SKILL.md`, required `name`/`description` frontmatter and relative resources.
+The [OpenAI skill guide](https://learn.chatgpt.com/docs/build-skills) distinguishes
+local skill folders from distributable plugins. Follow
+[plugin packaging](https://developers.openai.com/plugins/build/plugins) for
+repository marketplaces and host-specific metadata. Reviewed 4 November 2026.
+These are packaging contracts; host behavior still needs its own rehearsal.
+
+- [plugin.json](../../plugins/n8r-surface/plugin.json) uses Agent Plugins 1.0.0.
+- [.codex-plugin/plugin.json](../../plugins/n8r-surface/.codex-plugin/plugin.json)
+  is the compatibility manifest used by the verified Codex CLI 0.166.0. Keep
+  identity, version and presentation metadata aligned with the portable manifest.
+  Both use the same skill files; no second instruction copy is maintained.
+- [The repository marketplace](../../.agents/plugins/marketplace.json) exposes
+  `n8r-surface@n8r-surface` as available, without forcing installation.
+- Plugin and skill version `0.1.11` identify this experimental skill distribution. Increment
+  them when changing distributed instructions; record the Git revision and
+  installed file digest with test results. This is separate from the npm version.
+
+Keep the skill self-contained: installation copies it away from this checkout.
+Bundled references must work there. Instructions use CC-BY-4.0; metadata, icons
+and tooling use Apache-2.0. The package includes the repository's licence texts
+and [licensing explanation](../../LICENSING.md).
+
+## Maintainer installation from a local candidate
+
+Use this route only when developing the plugin itself. Use Codex CLI 0.166.0;
+other host versions need their own verification. Record `codex --version`
+and `git rev-parse HEAD`. If this marketplace is already registered from GitHub,
+run `codex plugin marketplace remove n8r-surface` before changing its source to
+the local checkout. The CLI rejects that source change otherwise. This removes
+the marketplace registration/snapshot, not research notes. From this repository root:
+
+```sh
+codex plugin marketplace add . --json
+codex plugin add n8r-surface@n8r-surface --json
+codex plugin list --marketplace n8r-surface --json
+```
+
+Expected: each command exits 0. The installed list contains exactly one
+`n8r-surface@n8r-surface`, version `0.1.11`, installed and enabled. The installation
+path varies by computer. `alreadyAdded: true` is fine on a repeated marketplace
+registration. Do not register a second copy of the skill in `.agents/skills`.
+
+If the earlier 0.1.0 candidate is installed, remove its old package ID once:
+
+```sh
+codex plugin remove n8r-researcher@n8r-surface --json
+```
+
+The renamed package replaces that candidate; saved research stays in place.
+
+For the pending ChatGPT desktop acceptance trial, open Plugins and select the **n8r-surface** source. Verify
+**n8r-surface Researcher** is installed. Refresh/restart the app if the new marketplace
+or version is absent, as the packaging guide requires. Then start a fresh Work
+chat. Loading instructions in this development chat is not this acceptance test.
+
+The repository installation above is sufficient for a trial. A public Plugins
+Directory listing requires a separate
+[submission and publication](https://developers.openai.com/plugins/deploy/submission)
+decision; it is not necessary for this repository test.
+
+## LOAD-01: install, invoke, remove, reinstall
+
+Use the prepared installation above. Record the desktop app version and selected
+Work environment. The host must have access to the two synthetic documents
+[lumen-guide.md](../../examples/researcher/sources/lumen-guide.md) and
+[lumen-settings.md](../../examples/researcher/sources/lumen-settings.md), through
+attachments or explicitly selected local source access. A local path by itself
+is not access in a cloud chat.
+
+1. In a **fresh Work chat**, type `@` and select the plugin's
+   `n8r-surface:research` skill (display name **n8r-surface Researcher**). Ask:
+
+   > Research Lumen Desk from these two documents for an AI Surface Model
+   > description. Identify the components and configuration choices, show evidence
+   > and gaps, and ask the next question that would help me choose a setup.
+   > Save working research notes that I can reopen.
+
+   Expected: the skill identifies version `0.1.11`, reads the two actual documents,
+   cites evidence and saves a readable note outside the plugin cache. Independently
+   open that note. Both conflicting batch maxima (10 and 25) and the unknown model
+   version must survive; optional features must not become user selections. The
+   note is not a validated `n8r.research/0.1` artefact. A version announcement alone
+   does not prove the packaged instructions or sources were used; retain the
+   visible invocation/tool evidence and verify the resulting content.
+2. Record the note's path/download and content digest when available. In the
+   plugin browser select **Uninstall plugin**, or run:
+
+   ```sh
+   codex plugin remove n8r-surface@n8r-surface --json
+   codex plugin list --marketplace n8r-surface --json
+   ```
+
+   Expected: exit 0, no installed entry for this plugin. The marketplace may still
+   offer it for installation. Do not remove other skills or the saved research.
+3. Start another **fresh Work chat**. Check that this skill is absent from the
+   selector and installed-plugin list. Do not paste its instructions into that
+   chat. Confirm the saved note still opens unchanged. Do not use an old chat's
+   memory of the skill as an unload test: uninstall does not erase its history.
+4. Reinstall using the same `codex plugin add` command. In a new Work chat, verify
+   exactly one skill entry appears and invoke it again against the same sources.
+   Supply the earlier note when testing continuation. Retain a new output file;
+   do not overwrite the first note. Verify version, source identity and retained
+   evidence. A prompt that never loaded the installed skill is not a pass.
+
+Any missing selector entry after install, duplicate installation, wrong version,
+fabricated source access, missing save/readback or changed user note is a failure
+or blocker to investigate. Do not count a CLI list result as a Work-mode invocation
+or unload result. Record the actual stage reached in [verification](researcher-skill-verification.md).
+
+For cleanup after testing, remove this plugin using the command above. Optionally
+run `codex plugin marketplace remove n8r-surface` to remove this marketplace source.
+Keep user research files. Reinstallation is reversible; cleanup is separate from
+publishing or deleting the source repository.
+
+## Prepare a shareable ZIP
+
+From this repository root with the pinned revision available locally:
+
+```sh
+n8r_skill_zip=$(mktemp -d)
+git archive --format=zip --output="$n8r_skill_zip/n8r-surface-0.1.11.zip" 5eed2ac9a31933caa2ae306e4b407e327d147bf7:plugins/n8r-surface
+python3 -m zipfile -l "$n8r_skill_zip/n8r-surface-0.1.11.zip"
+shasum -a 256 "$n8r_skill_zip/n8r-surface-0.1.11.zip"
+```
+
+Expected: exit 0; the archive has both manifests, one skill with its references, synthetic shape examples, workspace/artifact helpers
+and third-party notice
+and OpenAI metadata, the icon and licence files. It contains no Git history,
+source-repository research data, credentials or node_modules. The output directory
+varies. Archive only the committed plugin subtree so paused/untracked work cannot
+enter the package. Uploading or publishing the ZIP remains a separate action.
+
+## Installation rehearsal — 9 November 2026
+
+Codex CLI 0.166.0 on macOS installed the pinned merged revision from GitHub in a
+fresh temporary Codex home. A separate temporary home installed 0.1.8 at
+`3effc41bf7e8466651722b7b4c64035f862b1901`, then exercised the documented
+marketplace removal and pinned add/install/list sequence. All commands exited 0;
+each final list contained one installed/enabled 0.1.11 entry. All 20 installed
+files matched the pinned plugin subtree byte-for-byte. A synthetic research note
+outside the cache remained unchanged during replacement. The prepared ZIP's 20
+files also matched that subtree exactly.
+
+Raw command receipts and file comparisons remain in the private verification
+vault. The active host installation was unchanged. This verifies repository
+installation, replacement and ZIP preparation; it adds no inference, desktop
+lifecycle or multi-component research acceptance. The accepted research evidence
+remains the [8 November trial](skill-usefulness-verification.md).
+
+## Checked artifacts in 0.1.11
+
+The skill now supports standalone Capability/Bound JSON, research HTML and a
+brief through bundled resources. See the [artifact task and human
+verification](skill-surface-artifacts.md) for prerequisites, privacy limits,
+revision behavior and evidence. The [accepted two-product trial](skill-usefulness-verification.md)
+covers Capability, Bound and fresh-session revision on bounded LibreChat and Open
+WebUI source packs. It does not establish a live deployment or the proposed
+incremental LibreChat + Atlas + inference-service demonstration.

package/docs/n8r-surface/researcher-vault.md

@@ -0,0 +1,134 @@
+# Local AI Surface vault
+
+The vault is a local working directory for AI Surface research. Keep research
+notes, intermediate artifacts and saved Surface descriptions here. Its location
+is chosen on each machine and can sit outside the `n8r-surface` repository.
+
+## Where things go
+
+| Folder inside the vault | Contents |
+| --- | --- |
+| `research/<source-slug>/` | Numbered research notes and continued investigations. |
+| `artifacts/<source-slug>/` | Intermediate outputs, drafts and supporting evidence. |
+| `surfaces/<source-slug>/` | Saved Surface descriptions and their revisions. |
+| `examples/` | Optional local copies of the prepared test source material. |
+| `runs/<run-name>/` | Disposable test inputs, saved notes and review copies. |
+
+Use a short source name such as `example-chat`. A research note might live at
+`vault/research/example-chat/graph-001.md`. Keep earlier notes and revisions;
+use the next unused number when saving another. The agent should resolve and
+show the actual absolute destination before writing, then reopen the saved file.
+
+Within the `n8r-surface` checkout, [agent guidance](../../AGENTS.md#local-research-workspace)
+reads `.local/vault-path`: an ignored text file containing one absolute directory
+path. This is a local instruction convention, not a CLI or plugin configuration
+interface. The path is not published with the repository. When the vault is moved,
+update that local pointer and preserve the files; do not migrate existing research
+elsewhere without a separate request.
+
+Explicit destinations take precedence. Continue existing notes in their current
+folder unless the user asks to move them. Without a selected vault or another
+destination, the installed skill still defaults to
+`~/Documents/n8r-surface/research/<source-slug>/`. Installation does not create
+or populate a vault.
+
+The folder layout does not validate its contents. Current Researcher 0.1.7
+produces Markdown working notes; `surfaces/` holds Surface descriptions when
+available, without claiming the skill generates validated definitions. Local
+files are not automatically published, backed up or encrypted. If you place a
+vault in a Git repository, ignore its working areas before saving private work.
+
+## Try the installed skill
+
+The shared material below assumes **n8r-surface Researcher 0.1.7 is already
+installed**. Anyone with this checkout has the same
+[prepared source pack](../../examples/researcher-vault/README.md).
+
+Choose a walkthrough:
+
+- [Try the prepared example](researcher-prepared-check.md): use a small fictional chat
+  system, save its description, change one setting and compare the findings.
+  The source material is provided in this repository.
+- [Try another chat-system repository](researcher-repository-check.md): choose a real
+  project Researcher has not already investigated for you, then inspect its AI
+  connections and continue the investigation in a fresh chat.
+
+Use a local chat app that can read your files and save notes. Both walkthroughs
+assume you can select `n8r-surface:research`; [installation help](researcher-skill.md)
+is separate. No running chat service is required.
+
+## Start with fresh context
+
+For an independent attempt, start a new chat rather than continuing or forking
+the development chat. Before the first test prompt, use `/memories` to turn off
+both using and generating memories, where those controls are available. See
+[the app's memory controls](https://learn.chatgpt.com/docs/customization/memories).
+
+Read the walkthrough yourself and paste only its task prompts into the test
+chat. Supply the named inputs, not this checklist or its expected answers. For a
+continuation check, add only the previous note requested by that step. Ask the
+agent to use the installed skill and those sources, without consulting other
+chats, projects or the web, and inspect its source references and tool activity.
+
+This limits the intended evidence; it does not enforce isolation. A fresh chat
+still has applicable instructions and the model's general knowledge. Codex
+[permission profiles](https://learn.chatgpt.com/docs/permissions#file-access-limited-to-workspace)
+can restrict command reads to selected folders, but connectors, browser access
+and other tools have separate controls. Read-only mode alone is not a guarantee
+that neighboring repositories or expected answers cannot be read. These walkthroughs
+are practical manual checks, not sandboxed evaluations.
+
+## Keep or clean up a test
+
+Keep the repository's prepared examples unchanged so another person can start
+the same test.
+Keep each run until you have reviewed its findings. To start again, use a new run
+name. After review, you can delete that one folder under `runs/` with your file
+browser. Leave the shared examples, ongoing research, saved Surfaces and source
+repositories intact.
+
+## Standalone command setup
+
+The experimental bare `n8r-surface` command offers a local vault once and remembers
+it as `vaultRoot` in version-1 user settings. It creates `research/`, `artifacts/`
+and `surfaces/` only after confirmation, and reuses that vault from other source
+folders. The legacy `artefactRoot` remains separate and unchanged. This setup-only
+increment does not run the Researcher or save findings. It does not read the
+repository's `.local/vault-path`. See [startup behavior](startup.md) and the
+[installed verification tasks](vault-setup-verification.md).
+
+## Standalone saved continuation
+
+The command now writes `research/<source-slug>/research-NNN.md` only when the user
+enters `/save`. A successor stays beside its predecessor with the next unused
+number. The host appends experimental provenance metadata; keep that block intact
+when editing prose. Edits before or after the block are included on reopening.
+These notes are Markdown working research, not validated HTML or Surface JSON/YAML.
+
+Use `n8r-surface reopen NOTE --source SOURCE` to select one explicitly. Without
+NOTE, `reopen --source SOURCE` suggests the most recently filesystem-edited
+numbered note in that source's research folder. It shows the path and UTC edit time
+and asks before selecting it. Tied edit times use filename order. The source slug
+is a convenience only: the chosen source and saved evidence are compared anew.
+An invalid newest note reports an error rather than silently selecting an older one.
+
+Explicitly named notes may live outside the configured vault; saving continues in
+that note's folder. Only notes with this command's metadata envelope are supported
+in this increment; arbitrary older skill notes need a later import decision.
+The source is always selected explicitly again. See [the short check](getting-started.md)
+and [storage limits](save-reopen-verification.md).
+
+## Candidate structured Surface pairs
+
+Ask the conversation to create a Surface, inspect its validated preview, then use
+`/save surface`. Initial pairs go under
+`surfaces/<source-slug>/surface-NNN/`, containing `research.html`, `surface.json`
+and a host `manifest.json`. These are the experimental library's research and
+Surface contracts; the manifest is separate storage metadata with byte hashes and
+provenance. Keep all three files together when moving a pair.
+
+Reopen either HTML or JSON with `--source SOURCE` to start fresh and revise it.
+Successors stay beside the selected pair, including after relocation, and retain
+previous bytes. Omitted filenames still suggest Markdown notes only. Manual HTML
+or JSON edits conflict with the saved manifest; make revisions through the
+conversation. See [Surface evidence and limits](surface-revisions-verification.md).

package/docs/n8r-surface/save-reopen-transcript.md

@@ -0,0 +1,211 @@
+# Saved continuation: actual terminal transcripts
+
+Sanitized agent rehearsal on 5 November 2026 using installed candidate code
+`447f440`, Codex 0.166.0 and `gpt-6-astra`. Local paths are replaced with `<run>`;
+trailing whitespace is trimmed and model wording is retained. See [evidence and limits](save-reopen-verification.md).
+
+## Research and first save
+
+````text
+Research source: <run>/source/lumen-guide.md (one file)
+Inspected source text and your messages will be sent to OpenAI through Codex using ChatGPT sign-in.
+Source access is read-only. /save explicitly saves a numbered Markdown working note.
+Start research with OpenAI using Codex ChatGPT sign-in? [y/N] y
+
+Connecting to Codex…
+Connected: ChatGPT sign-in; model gpt-6-astra.
+Researcher 0.1.8; sha256:8af29c3073df9b8210a7dd94ced9700379b0a98f76bd3a2902f9430b93c98a2a
+Ask about your source. /save saves a note; /exit ends the chat; Ctrl-C cancels.
+You> Describe the AI components and connections with stable component IDs and source citations. For this investigation I choose not to use external OIDC: retain that as my decision, not a change to the source. Keep unknowns and useful next steps clear.
+
+I’m using `n8r-surface:research 0.1.8`. I’ll map the supplied guide and retain your choice not to use external OIDC separately from what the source describes.
+[source_list] .
+[source_read] lumen-guide.md sha256:98d3374c79c166b37f0debb4ffb5e457eda1385dd7847b042cbad674cac98693
+## Partial AI Surface: Lumen Desk
+
+**Scope:** Fictional guide `fixture-1`, not a real installation. These are documented capabilities, not verified runtime connections. [lumen-guide.md:1–4](lumen-guide.md#L1-L4)
+
+### Components and connections
+
+| Stable ID | Role and directed connections | Evidence |
+|---|---|---|
+| `component:lumen` | Support-summary application → `component:inference` to generate summaries. Input sources and request path are unspecified. | [Lines 3–9](lumen-guide.md#L3-L9) |
+| `component:inference` | Configured inference endpoint; underlying model, version and service identity are unknown. | [Lines 8–9](lumen-guide.md#L8-L9) |
+| `component:archive` | `component:lumen` → archive service through **separate HTTP routes** for summary documents and audit events. These share one service, not two components. Documented batch limit: 10 documents. Delivery is an option, not an established deployment selection. | [Lines 11–16](lumen-guide.md#L11-L16) |
+| `component:callback` | Optional receiver for `component:archive` → notification delivery. Disabled until selected and configured; receiver hosting is unspecified. | [Lines 18–22](lumen-guide.md#L18-L22) |
+| `component:oidc` | Supported external identity-service option for `component:lumen`; **not selected for this investigation**. | [Lines 8–9](lumen-guide.md#L8-L9) |
+
+### Your decision
+
+**2026-11-05:** Do not use external OIDC for this investigation. This is your scoped selection—not a source edit, proof that OIDC is disabled in a deployment, or evidence of an alternative authentication mechanism.
+
+### Unknowns and useful next steps
+
+- **Inference:** Identify the endpoint, model/version and submitted data. Resume at lines 8–9 with applicable nonsecret configuration or implementation evidence.
+- **Archive:** Establish whether delivery is selected, then clarify request schemas and authentication for both routes. The guide explicitly does **not** establish a tested Harbor Archive integration. Resume at lines 13–16.
+- **Callbacks:** Confirm whether you want this option. If selected, investigate notification validation, retry behavior and protocol; all remain unspecified. Resume at lines 20–22.
+- **Identity without external OIDC:** Determine what access controls, if any, apply to Lumen and its connections. The guide provides no alternative identity design.
+
+**Coverage and identity:** Read all 29 lines of the sole supplied file; no exclusions or truncation. No deployment or runtime observations were supplied. [Lines 24–28](lumen-guide.md#L24-L28)
+Source SHA-256: `98d3374c79c166b37f0debb4ffb5e457eda1385dd7847b042cbad674cac98693`.
+
+To retain this as a numbered working draft, proposed name `surface-draft-001.md`, use **`/save`**. Nothing has been saved yet.
+You> /save
+
+Save destination: <run>/vault/research/lumen-guide-md/research-001.md
+Staged the working note with stable component IDs, citations, source hash, your OIDC decision, unknowns and next steps. The host will handle saving and verification.
+Saved and read back: <run>/vault/research/lumen-guide-md/research-001.md
+sha256:9910959c740959e943e7d6cf4e81b1d365468ab33576fb9c83aa9ce25dc054e7
+You> /exit
+````
+
+## Fresh reopen and successor
+
+````text
+Research source: <run>/source/lumen-guide.md (one file)
+Inspected source text and your messages will be sent to OpenAI through Codex using ChatGPT sign-in.
+Source access is read-only. /save explicitly saves a numbered Markdown working note.
+Saved note: <run>/vault/research/lumen-guide-md/research-001.md
+Last edited: 2026-11-05T11:08:20.894Z
+Recorded source: lumen-guide.md (file); this is not verified identity.
+The saved note, including human edits, will also be sent to OpenAI after confirmation.
+Start research with OpenAI using Codex ChatGPT sign-in? [y/N] y
+
+Connecting to Codex…
+Connected: ChatGPT sign-in; model gpt-6-astra.
+Researcher 0.1.8; sha256:8af29c3073df9b8210a7dd94ced9700379b0a98f76bd3a2902f9430b93c98a2a
+Ask about your source. /save saves a note; /exit ends the chat; Ctrl-C cancels.
+[reopen evidence] lumen-guide.md: unchanged sha256:98d3374c79c166b37f0debb4ffb5e457eda1385dd7847b042cbad674cac98693
+I’m continuing with `n8r-surface:research 0.1.8`. I’ll re-read the selected source, compare it with the historical claims, and retain your OIDC decision separately. I won’t save anything.
+
+[source_list] .
+[source_read] lumen-guide.md sha256:98d3374c79c166b37f0debb4ffb5e457eda1385dd7847b042cbad674cac98693
+## Continued partial Surface
+
+The freshly read source identifies itself as **“Lumen Desk — fictional guide, fixture-1”**, describing a synthetic application—not a real installation (`lumen-guide.md:1–4`).
+
+| Directed connection | Current evidence |
+|---|---|
+| `component:lumen` → `component:inference` | Summary generation through a configured endpoint; service identity, model and version remain unspecified (`lumen-guide.md:8–9`). |
+| `component:lumen` → `component:archive` | Optional document and audit delivery through separate HTTP routes on **one service**. Maximum 10 documents per batch; schema/authentication unresolved; no tested Harbor integration (`lumen-guide.md:13–16`). |
+| `component:archive` → `component:callback` | Optional notifications; receiver disabled until selected and configured. Validation, retries and tested protocol remain unspecified (`lumen-guide.md:20–22`). |
+| `component:oidc` → `component:lumen` | Supported external identity option, **excluded from your intended selection** by the retained human decision—not evidence of runtime disablement (`lumen-guide.md:8–9`). |
+
+### Reconciliation and coverage
+
+- Freshly listed the selected source and read all 29 lines; no exclusions or truncation were reported.
+- Fresh SHA-256: `98d3374c79c166b37f0debb4ffb5e457eda1385dd7847b042cbad674cac98693`. It matches the predecessor’s recorded digest independently of the supplied comparison.
+- No changed source claims were found. Prior findings remain historical; the table above is freshly corroborated documentary evidence. Matching bytes do **not** establish deployment identity or exercised connections.
+- Retained all component IDs and the dated decision not to use external OIDC. No separately identified human edits were supplied.
+- Implementation, deployment configuration and runtime evidence remain unavailable; the guide explicitly supplies no deployment or runtime observations (`lumen-guide.md:26–28`).
+
+### Next useful investigation
+
+Prioritize `component:inference`: identify its selected endpoint/service, model/version and submitted summary data. Resume from `lumen-guide.md:8–9` with explicitly authorized, nonsecret configuration or implementation evidence. This would clarify the main AI data boundary; the current guide cannot resolve it.
+
+Can you supply nonsecret inference configuration or implementation evidence as the next selected source?
+
+Nothing was saved.
+You> Continue the authentication investigation. What remains unknown, and what evidence would resolve it? Retain my earlier decision.
+
+## Authentication: remaining gaps
+
+**Retained decision (2026-11-05):** You chose **not to use external OIDC for this investigation**. `component:oidc` remains a supported option, excluded from your intended selection—not proven disabled in a running system.
+
+The guide only establishes OIDC support (`lumen-guide.md:8–9`); it does not identify an alternative authentication mechanism.
+
+| Unknown | Evidence that would resolve it |
+|---|---|
+| How users authenticate to `component:lumen` without external OIDC—or whether access is anonymous | Authentication documentation, login handlers and request-authentication middleware. |
+| Which authentication mode is actually selected | Applicable nonsecret deployment configuration and configuration-loading/override logic. Runtime-effective evidence would be needed to establish active settings. |
+| What authenticated users may access or do | Authorization checks, role definitions and ownership checks governing summary generation and access to summaries. |
+| How sessions are managed, if sessions exist | Session implementation and nonsecret settings for expiry, revocation and cookie protections. |
+| How `component:lumen` authenticates to `component:inference` | Inference client implementation and documented authentication scheme; no credential values needed. |
+| How archive delivery is authenticated, if selected | Selected archive interface documentation and client configuration. The guide explicitly leaves authentication to separate agreement (`lumen-guide.md:13–15`). |
+| How callbacks authenticate their sender, if selected | Receiver implementation and protocol documentation covering signature/token verification and replay protection. Validation is unspecified (`lumen-guide.md:20–22`). |
+
+These are **unknowns, not established absences or vulnerabilities**. The selected source contains no implementation or runtime observations (`lumen-guide.md:26–28`).
+
+**Next useful investigation:** Trace the incoming request authentication and authorization checks for `component:lumen`. This would establish what protects summary generation without external OIDC.
+
+Can you provide authentication documentation or request-handler code as the next explicitly selected source?
+You> /save
+
+Save destination: <run>/vault/research/lumen-guide-md/research-002.md
+Staged the working note with the graph, source hash, retained OIDC decision, authentication gaps and next steps. Host saving and verification remain pending.
+Saved and read back: <run>/vault/research/lumen-guide-md/research-002.md
+sha256:927a6a84a810d518b5e7bc8f37df3f29916e353a3bb6817ccf21b2cc36aee265
+You> /exit
+````
+
+## Latest-edited suggestion and human edit
+
+````text
+Research source: <run>/source/lumen-guide.md (one file)
+Inspected source text and your messages will be sent to OpenAI through Codex using ChatGPT sign-in.
+Source access is read-only. /save explicitly saves a numbered Markdown working note.
+Saved note: <run>/vault/research/lumen-guide-md/research-001.md
+Last edited: 2026-11-05T11:09:49.624Z
+Recorded source: lumen-guide.md (file); this is not verified identity.
+Open this note? [y/N] y
+
+The saved note, including human edits, will also be sent to OpenAI after confirmation.
+Start research with OpenAI using Codex ChatGPT sign-in? [y/N] y
+
+Connecting to Codex…
+Connected: ChatGPT sign-in; model gpt-6-astra.
+Researcher 0.1.8; sha256:8af29c3073df9b8210a7dd94ced9700379b0a98f76bd3a2902f9430b93c98a2a
+Ask about your source. /save saves a note; /exit ends the chat; Ctrl-C cancels.
+[reopen evidence] lumen-guide.md: unchanged sha256:98d3374c79c166b37f0debb4ffb5e457eda1385dd7847b042cbad674cac98693
+I’m continuing with `n8r-surface:research 0.1.8`, prioritizing the human edit about session expiry and retaining the external OIDC non-selection. I’ll re-read the selected source; I won’t save anything.
+
+[source_list] .
+[source_read] lumen-guide.md sha256:98d3374c79c166b37f0debb4ffb5e457eda1385dd7847b042cbad674cac98693
+## Continued partial Surface
+
+**Fresh source check:** Read all 29 lines of the selected `lumen-guide.md`; listing and read reported no exclusions or truncation. Its title and scope identify a fictional fixture, not a real installation (`lumen-guide.md:1–4`). SHA-256 matches the prior note:
+
+`98d3374c79c166b37f0debb4ffb5e457eda1385dd7847b042cbad674cac98693`
+
+This establishes unchanged source bytes—not deployment or system identity. The supplied comparison remains historical input; the fresh read independently supports byte agreement.
+
+### Retained graph
+
+These are documented capabilities, not exercised connections.
+
+| Directed connection | Current evidence |
+|---|---|
+| `component:lumen` → `component:inference`: summary generation | Model and version unspecified (`lumen-guide.md:8–9`). |
+| `component:lumen` → `component:archive`: optional document/audit delivery | Separate routes on one service; schema/authentication unresolved; maximum 10 documents per batch; not a tested Harbor integration (`lumen-guide.md:13–16`). |
+| `component:archive` → `component:callback`: optional notifications | Disabled until selected/configured; validation and retries unspecified (`lumen-guide.md:20–22`). |
+| `component:lumen` → `component:oidc`: supported identity option | Documented support remains (`lumen-guide.md:9`); **human non-selection retained**, not an observed runtime setting. |
+
+### Session-expiry priority
+
+The human edit now takes priority over the predecessor’s inference-endpoint investigation. The complete source contains **no session-expiry evidence**: no session lifetime, idle timeout, renewal, logout invalidation, or enforcement behavior is described. This is missing evidence, not evidence that sessions never expire. No alternative authentication mechanism is established.
+
+Prior claims remain preserved as historical findings and are consistent with the fresh source. No changed source evidence was found; deployment and runtime evidence remain unavailable (`lumen-guide.md:24–28`).
+
+**Next useful investigation:** establish how `component:lumen` enforces session expiry without external OIDC, using explicitly authorized authentication documentation or implementation and applicable nonsecret configuration.
+
+Can you supply session-management documentation or code for that next check?
+
+Nothing saved.
+You> Which human edit and decision did you retain? Continue from them without inventing source facts.
+
+I retained:
+
+- **Human edit:** “prioritize session-expiry evidence next.”
+- **Human decision:** do **not** select external OIDC for this investigation. This does not establish that OIDC is disabled in a running system.
+
+Continuing from those: session expiry is the next investigation priority for `component:lumen`. The inspected guide documents external OIDC as an available option (`lumen-guide.md:8–9`), but supplies no session-expiry or alternative authentication details. It also supplies no deployment or runtime observations (`lumen-guide.md:24–28`). Session behavior therefore remains **unknown**.
+
+Can you provide authorized session-management documentation or implementation evidence covering expiry without external OIDC?
+You> /save
+
+Save destination: <run>/vault/research/lumen-guide-md/research-003.md
+Staged the continuation draft, retaining the graph, source hash, OIDC non-selection, and session-expiry priority. The host will handle saving and verification.
+Saved and read back: <run>/vault/research/lumen-guide-md/research-003.md
+sha256:56f317af8dfd30ea76cbfe0318bf74b226cc3be0ef851e0edc67bcd120d029d6
+You> /exit
+````