@n8r/surface 0.1.0-experimental.2 → 0.1.0-experimental.5

package/docs/n8r-surface/skill-bom-verification.md

@@ -0,0 +1,142 @@
+# Self-contained Surface verification
+
+Issue [#68](https://github.com/injectionator/n8r-surface/issues/68), draft
+[PR #69](https://github.com/injectionator/n8r-surface/pull/69). Candidate 0.1.13.
+This record separates exercised behavior from pending human acceptance.
+
+## Baseline
+
+An actual installed 0.1.12 session inspected its instructions/helpers and ran
+exporter help. It reported that BOM-only recovery was unavailable: native records
+and mapping siblings were required. This was a capability inspection, not a full
+research round trip. No product findings or historical research were supplied.
+
+## Synthetic checks
+
+The new fixture retains four database alternatives, one selected dependency,
+unknown inference, source evidence, a decision and investigation entries in one
+BOM. Tests exercise copied-BOM recovery, selected graph equality, source/evidence
+references, independent observations, hosting distinctions, successor preservation,
+changed-view reconciliation, unsafe directories and HTML escaping. These are
+bounded automated checks, not model conformance or human acceptance.
+
+Sunshine commit `beaab3faf2bab234069a8677fc627de2f0f01a3d` generated HTML with the
+synthetic application, model and services but omitted profile alternatives/evidence
+and investigation entries. It is not the selected investigation renderer.
+
+Browser review of the generated local HTML was blocked by the browser tool's
+`file:` URL policy. No alternate route was attempted. File-level checks do not
+establish visual/browser acceptance.
+
+CycloneDX Python Library 11.12.0 with JSON validation extras accepted the synthetic
+BOM against 1.7 and loaded both components, four services and sixteen properties.
+It exposes no `annotations` attribute in this deserialization API; no complete
+profile round-trip claim is made. The native profile reader checks the complete
+BOM directly, including annotations and references.
+
+## Independent review
+
+Three material findings were reproduced in tests and fixed: BOM-path bypass of
+edited sibling views, disconnected inventory missing from human views, and dropped
+historical observations. The focused suite now has fifteen passing checks.
+
+## Installed assessment
+
+The user explicitly selected a local LibreChat checkout. Sources, transcripts and
+research remain in a new private run, separate from prior historical rehearsal.
+The everyday installation remains unchanged. All sessions used Codex CLI 0.166.0,
+ChatGPT sign-in, configured `gpt-6-astra`, Node 26.13.0 and macOS. No API-key or
+provider/model fallback. The exact initial installation matches `2b2816e`; the
+reviewed candidate used for diversion/continuation matches
+`bd2b8aef4ef7e6154d40c23349a846be69146da8` (24 plugin files).
+
+The harness supplied the installed skill, selected source and writable private
+workspace. Network, memory, unrelated tools and repository instruction loading
+were disabled. Filesystem probes checked denied historical evidence; these
+specific controls are not a general claim of host-memory isolation.
+
+The installed Researcher saved revision 001 with an initial plan before research.
+It independently inspected bounded ranges in fourteen files and saved revision
+002 with twelve nodes, thirteen relationships, twelve evidence records, ten
+journal entries and five gaps. All fourteen source hashes matched independent
+readback. No deployment selections were asserted. Git inspection was blocked
+inside the rehearsal; the Researcher recorded inspected hashes and the limit,
+rather than asserting a clean checkout. Developer-only source identity metadata
+is retained privately and was not supplied as research findings.
+
+The scripted user diversion paused tool/document investigations and prioritized
+authentication. The reviewed candidate inspected two additional files and saved
+revision 003. The BOM retains the rehearsal user decision, seventeen journal
+entries and three deferred paths with reasons. All ten files in revisions 001/002
+remain byte-identical. This is an agent-run user scenario, not human acceptance.
+
+One independent presentation assertion searched for an unescaped setting name in
+Markdown and failed after the successful save. The installed agent normalized
+Markdown-escaped underscores and completed readback; both results are retained.
+No saved files were replaced to make the assertion pass.
+
+For the final continuation, only revision 003's BOM was copied to a new writable
+workspace. The harness confirmed that the old workspace was denied and the prior
+ephemeral thread was unavailable. The reviewed installation recovered the objective,
+authentication-first choice, deferred paths/reasons and uncertainties, then saved
+revision 004 with a brief/map/journal. It did not need native siblings or old chats.
+
+Independent readback found twelve nodes, thirteen unchanged relationships, fifteen
+evidence records, twenty-two journal entries, two scoped rehearsal decisions and
+three deferred paths. All inherited evidence and source-backed claims were unchanged;
+the additional user-basis record describes continuation scope, not a source finding.
+All fifteen files in revisions 001–003 remain byte-identical. Seventeen source
+references across sixteen unique inspected files match their hashes; the original
+checkout's revision and working-tree status remain unchanged.
+
+The developer's first final assertion mistakenly required equality of the whole
+evidence array, rejecting the legitimate appended continuation-scope record. It
+was corrected to require preservation of inherited evidence and equality of all
+non-user source evidence; that check passed. Both attempts are recorded privately.
+
+These are agent rehearsals of the documented interactions. They do not establish
+human acceptance, actual deployment selections, runtime effectiveness or exhaustive
+source coverage. Selected/possible visual semantics are exercised by the synthetic
+fixture; the real assessment appropriately leaves deployment selection unknown.
+
+## Human task list
+
+Prerequisites: candidate 0.1.13 installed, Node 22/24/26 and readable selected
+sources; a new writable Surface folder outside the source checkout. The maintainer
+provides the exact candidate revision/digest and rehearsal output separately.
+
+- [ ] Open a fresh chat with the installed Researcher and the selected source.
+  Ask: “Explain the AI-relevant paths in this source. Save the Surface and maintain
+  an investigation journal in this folder as you go. Show me what to inspect next.”
+  Expect an initial plan, useful partial findings and openable journal/map/brief.
+- [ ] Inspect the map. Possible paths and selected paths must look different.
+  Open a relationship's evidence; unknown selection must not be called absent.
+  Selection must not be called observed runtime use without dated evidence.
+- [ ] Redirect: “Pause that path and investigate authentication first. Keep the
+  deferred work and explain what access would help.” Expect a recorded diversion,
+  deferred reason, new next steps and a new revision preserving earlier files.
+- [ ] In another fresh chat supply only the latest BOM and source access. Ask:
+  “Continue this Surface. What did I choose, what remains uncertain, and what
+  should we investigate next?” Expect recovery of the choices, evidence and
+  investigation state without requiring a native sibling or old chat.
+- [ ] Report **Pass / Fail / Blocked**, the interaction, candidate revision and
+  relevant output paths. Fail for lost decisions, invented bindings, changed old
+  files or missing state; Blocked for inaccessible sources/runtime. Keep source
+  data private. Retain outputs until reviewed; cleanup is only the disposable run.
+
+## Automated checks and retained candidate
+
+- Node 22.23.2, 24.22.2 and 26.13.0: 224 tests and package checks pass on each.
+- Markdown lint 0.24.0, offline lychee 0.24.2 links/anchors, cffconvert 2.0.0,
+  JSON parsing (18 examples), workflow shell syntax and 24 Python tests pass.
+- `npm ci` and `npm cache add yaml@2.9.1` completed; no new runtime dependency.
+- Exact reviewed candidate ZIP retained privately; SHA-256
+  `bdab66880bc1763c73ee30d98d3b188a4c8ae2af410f03b364785fbcc58640dd`.
+  No everyday plugin replacement, release or merge.
+
+The first package check failed because its golden instruction digest still named
+the earlier instructions; the reviewed digest was updated and all three package
+checks passed. Earlier red regression runs are retained as development evidence,
+not counted as final successes. Browser visual review and human acceptance remain
+pending. Structure, byte integrity and named consumer reads do not prove research
+accuracy or deployment behavior.

package/docs/n8r-surface/skill-cyclonedx-verification.md

@@ -0,0 +1,144 @@
+# Historical skill 0.1.12 CycloneDX verification
+
+This alongside-records rehearsal used explicitly supplied historical notes and
+model intent; it was not a neutral fresh assessment. It does not verify the revised
+BOM-authority design. See [candidate 0.1.13 evidence](skill-bom-verification.md).
+
+Issue [#68](https://github.com/injectionator/n8r-surface/issues/68), draft
+[PR #69](https://github.com/injectionator/n8r-surface/pull/69). Agent rehearsal on
+13 November 2026; human acceptance, merge and release remain separate decisions.
+
+## Revision and environment
+
+The isolated installation contains all 22 plugin files exactly matching source
+commit `da0e445221d2209fc843d0cef682606610d22ad5`, candidate 0.1.12.
+Codex CLI 0.166.0, ChatGPT sign-in, configured `gpt-6-astra`, Node 26.13.0,
+macOS. No API-key or provider/model fallback. The active everyday installation
+remained 0.1.11; this candidate used a separate host home and private workspace.
+
+The harness uses ephemeral app-server sessions with the installed skill supplied
+as a skill input. Network, memory, unrelated tools and repository instructions are
+disabled. Only the selected workspace and installed skill are accessible, apart
+from required runtime files. A filesystem probe checks that private evidence and
+unsupplied repository instructions cannot be read. Source-pack provenance, raw
+transcripts, outputs, hashes and environment receipts stay in the selected private
+vault; this public record contains no private deployment configuration.
+
+## Baseline and task
+
+State X: installed 0.1.11, supplied source pack and writable workspace. Input Y:
+inspect the installed skill/helper and determine whether checked CycloneDX export
+is available, without substituting invented findings or an external exporter.
+Observed: only native artifact operations were available; the host identified the
+missing mapping/export/validation workflow. This was a bounded capability check,
+not a failed full research run.
+
+The candidate task was to research the supplied multi-component sources, explain
+the system to a consultant advising a security engineer, preserve intended versus
+historical configuration, save native records and CycloneDX alongside them, and
+read both back. The source pack contained selected code, documentation and safe
+historical operational projections recovered from prior work. No applicable SBOM
+was supplied; network lookup was unavailable. Upstream revision, package and
+transitive coverage remain explicit gaps.
+
+## Observed initial result
+
+The installed candidate performed its own investigation and preparation. The
+developing agent supplied sources and user instructions, not prewritten findings.
+
+- Twelve recorded source hashes match their supplied files after the run.
+- Checked native Capability records contain nine nodes and nine relationships.
+- CycloneDX 1.7 contains five components and four services. It has no concrete
+  dependencies because the candidate chose Capability for this evidence scope.
+- Native snapshot equals the checked native bundle. Mapping/report retains the
+  connection meanings, evidence and uncertainty absent from the BOM view.
+- Intended model and different historical model evidence remain distinct. Source
+  control points are identified without asserting present deployment or efficacy.
+- Native brief, HTML and JSON, consultant summary and five-member exchange were
+  saved and independently reopened. The candidate corrected one inspection error
+  (`mapping.nodes` is nested inside the report), then completed readback. It did
+  not replace the successfully checked export with unchecked JSON.
+
+## Fresh-session correction
+
+A second ephemeral process confirmed that the previous thread was unavailable,
+then recovered from the saved directories and supplied sources. The rehearsal
+input added an owner decision to keep connector tools disabled in the intended
+configuration and require separate authorization/assessment for a future enabled
+path. It was explicitly design intent, not evidence of deployed configuration.
+
+The candidate saved native and exchange successors. Independent developer readback
+confirmed all ten earlier artifact files and all twelve sources were unchanged;
+all nine node identities and the prior decision were preserved. Decisions grew
+from one to two. The BOM serial stayed stable and its version advanced from 1 to
+2. Native snapshots matched both checked bundles. This is an actual installed-agent
+rehearsal, not human acceptance.
+
+## Independent consumer
+
+CycloneDX Python Library **11.12.0**, Python 3.13.1, in a separate environment:
+`JsonStrictValidator(SchemaVersion.V1_7)` returned no validation error;
+`Bom.from_json(data=json.loads(text))` deserialized the synthetic configured BOM.
+The consumer listed its application and separate services and resolved the
+selected application dependency to the concrete Atlas target rather than its
+abstract database role. The [retained synthetic exchange](../../examples/researcher/cyclonedx/configured/README.md)
+is available for inspection. The same independent consumer then validated and
+loaded both actual installed-skill BOMs: five components, four services, versions
+1 and 2, no dependency edges, as expected for this Capability projection.
+This establishes those bounded read tasks only; it
+is not lossless round-trip, general consumer or runtime compatibility.
+
+## Maintainer checks
+
+- Full suite: 208 passing on Node 22.23.2; 209 passing on Node 24.22.2 and
+  26.13.0 after adding the installation-buffer regression. Its focused 15-test
+  suite also passes on Node 22.23.2.
+- Installed package checks pass on all three Node majors, including existing
+  terminal setup/reuse/preservation interactions. Exact skill bundle and notices
+  checks pass; no runtime dependency installation is needed by either helper.
+- Twenty-four Git workflow tests and shell syntax check pass. Example JSON parses;
+  native contracts/reference checks and CycloneDX schema/reference checks are
+  bounded validation, not universal model conformance.
+- Markdown lint 0.24.0, offline links with lychee 0.24.2, citation metadata with
+  cffconvert 2.0.0 and whitespace checks are run before the final handoff.
+
+Two integration failures were found and fixed during actual checks: installed-byte
+verification exceeded its old one-MiB Git output buffer, and package smoke checks
+still expected the previous instruction version/digest. The buffer now permits
+16 MiB with a large-helper regression; the digest was independently calculated
+from the candidate source bytes. A fresh whole-branch reviewer confirmed the
+package-pin failure and found no critical projection issue. Additional tamper and
+classification boundaries were manually exercised by that reviewer; retaining
+more automated variants and clarifying the shared skill paragraph's CLI boundary
+remain minor follow-ups. The new exchange is a standalone skill capability, not
+a new product CLI command.
+
+## Human task list
+
+Prerequisites: candidate 0.1.12, Node 22/24/26, a readable source set and a writable
+disposable workspace. Open a fresh Codex chat in that workspace. Maintainers can
+install the pinned source through the normal plugin marketplace mechanism; see
+[skill installation](researcher-skill.md). Do not substitute an older installation.
+
+- [ ] Supply your source paths and say: `Use n8r-surface Researcher to explain this
+  multi-component Surface. Save checked native records and CycloneDX 1.7 alongside
+  them. Explain what the evidence establishes and what remains unknown.` Expect
+  source-backed findings and links to the native brief and exchange files.
+- [ ] Open the brief and `mapping.json`. Expect intended/historical/current state
+  to remain distinct. Open `surface.cdx.json`; expect identified components and
+  services, with only eligible explicitly selected dependencies. Capability output
+  may have none. Unknown coverage is explicit.
+- [ ] Start a fresh chat, supply saved directories and say: `Reopen these records.
+  Record this correction as a design decision, save successors and preserve all
+  earlier records and decisions.` Supply a correction. Expect stable node IDs and
+  BOM serial, incremented BOM version and links to both generations.
+- [ ] Report **Pass / Fail / Blocked**, candidate revision, interactions and output
+  paths. Missing output, invented facts, lost decisions or modified prior records
+  fail. Missing installation/source access blocks. Keep evidence for review; later
+  remove only the disposable workspace, not original sources.
+
+Values such as names, node counts and selected dependencies depend on the supplied
+sources; the nine-node rehearsal is not a required graph size. The
+[working manual](skill-manual.md) explains the user journey. Technical helper
+verification uses exit 0 for `saved-and-read-back` / `checked`; a nonzero exit is
+not success. Those commands are maintainer evidence, not required user interactions.

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

@@ -0,0 +1,85 @@
+# Self-contained CycloneDX Surface
+
+Skill **0.1.13** introduced the format below; candidate [0.1.14](first-use-overview.md)
+adds checked optional grouping and a connected overview without changing profile 1.
+The 0.1.13 increment replaces the 0.1.12 alongside-native exchange as the
+destination for new Surface saves. Issue [#68](https://github.com/injectionator/n8r-surface/issues/68).
+One CycloneDX 1.7 BOM carries inventory, relationships, evidence, decisions,
+uncertainties and investigation state. The brief, HTML map and Markdown journal
+are generated from it. Earlier saves remain intact and deliberately readable.
+The standalone product CLI retains its existing pair behavior; this is skill-first.
+
+## Standard fields and profile
+
+The [CycloneDX 1.7 reference](https://cyclonedx.org/docs/1.7/json/) supplies
+components/services, metadata, dependencies, annotations, external references,
+citations and properties. The [BOM-Link specification](https://cyclonedx.org/capabilities/bomlink/)
+provides version-pinned predecessor identity. We use one BOM per revision, not a
+set whose missing member could hide required investigation state.
+
+The exact [Surface profile](../../plugins/n8r-surface/skills/research/references/cyclonedx.md#profile-properties)
+and [synthetic example](../../examples/researcher/surface-bom/initial.cdx.json)
+are bundled with the installed skill. Profile version 1 uses unregistered
+`n8r:surface:` properties: objective, scope, individual relationship/evidence/entry
+records, coverage and gaps. It does not embed the native research/Surface bundle.
+
+| Meaning | Standard representation and bounded addition |
+| --- | --- |
+| Application, model, service | `components` / `services` with stable `bom-ref`; model type `machine-learning-model`. |
+| Selected dependency | Standard `dependencies`, mechanically checked against selected dependency-kind profile relationships. |
+| Supported alternative | Individual relationship property with support evidence, optionality/group, selection and conditions. Never an unconditional dependency merely because supported. |
+| Hosting, integration, data flow | Distinct relationship kinds, not interchangeable `dependsOn` edges. |
+| Dated observed use | Observation records with timestamp and observation-basis evidence; independent of support/selection. |
+| Findings, choices, disagreements | Standard annotations with subjects, annotator, timestamp and readable text. |
+| Source identity | External references with safe locator, hash, source version/capture scope in comment; profile evidence points to exact locations and applicability. |
+| Attribution | Standard citations with checked JSON Pointers and attribution references. Not a replacement for evidence. |
+| Investigation journey | Ordered entry properties: objective-driven steps, questions, findings, decisions, branches, supersession, access needs and reasons for deferral. |
+| Lifecycle | Capture context in metadata when established; never a forced whole-Surface evidence state. |
+| Revision | Stable serial, increasing version, exact predecessor BOM-Link and byte digest. |
+
+Missing configuration is unestablished selection, not absence. A placeholder is
+not a concrete binding. Safe static configuration can establish selection.
+Selection evidence distinguishes actual configuration from human design intent.
+Optionality is not a confidence score. A group alone does not assert exclusivity.
+
+## Save and recovery boundary
+
+The standalone `surface.mjs` checks official pinned schemas, profile shapes,
+references, evidence applicability and exact dependency projection. It saves one
+new private revision folder with BOM, generated views and a byte manifest.
+The BOM alone reopens; sibling files are not semantic dependencies.
+
+For directory continuation, edited generated views block ordinary save. Explicit
+reconciliation reads the proposals, retains the edited files, records an accepted
+choice/explanation and writes a successor. Predecessor BOM hashes are still checked.
+Supplying a lone BOM proves only its own structure and supplied predecessor identity,
+not that an unavailable old manifest or source is intact. No authenticity claim.
+
+The reader preserves historical annotations/source references and decision
+provenance; corrections add evidence and can supersede decisions. Nodes and
+relationship endpoints/kinds retain identity. Older revisions are never rewritten.
+Unsupported profile versions, dangling references and ambiguous projections fail.
+
+Legacy `artifacts.mjs read` and `cyclonedx.mjs read` remain available for deliberately
+supplied old saves. Migration requires reviewing and transferring their meanings
+into a fresh profile BOM, with explicit provenance/limits. No automatic lossless
+legacy conversion is claimed. No second native authority is created for new saves.
+
+## Consumer and visualization limits
+
+Generic consumers can read standard inventory and selected dependencies. They
+cannot reconstruct possible paths or investigation state without this profile.
+The helper renders those distinctions as solid selected and dashed other paths,
+with support state, evidence, observations and journal available for inspection.
+
+Sunshine at commit `beaab3faf2bab234069a8677fc627de2f0f01a3d` generated HTML from
+the synthetic 1.7 BOM, including application/model names and all four services.
+Its output omitted the profile relationship IDs, evidence and decision entries.
+It is therefore not selected as the investigation visual. This is a file-level
+consumer exercise, not a browser rendering or general 1.7 compatibility claim.
+See [Sunshine source](https://github.com/CycloneDX/Sunshine/tree/beaab3faf2bab234069a8677fc627de2f0f01a3d).
+
+The visual requires no scripts, remote assets or network. HTML escaping and a
+restrictive content policy keep saved text inert. Browser visual QA is separately
+tracked in the [verification record](skill-bom-verification.md). Validity does not
+establish research accuracy, complete inventory, deployed controls or runtime use.

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

@@ -0,0 +1,129 @@
+# Researcher skill: working user manual
+
+For installation and a short first trial, use the [tester guide](tester-candidate.md).
+
+Candidate 0.1.13 saves the Surface with a brief, visual map and investigation
+journal. This manual develops through actual use; the
+[verification record](skill-bom-verification.md) distinguishes exercised behavior
+from pending human acceptance. The earlier [0.1.12 rehearsal](skill-cyclonedx-verification.md)
+used historical material and a different save format; it was not a neutral fresh
+assessment.
+
+## Start with what you have
+
+Candidate 0.1.16 can begin with a short description or deployment document.
+Researcher should show a provisional map, then help you provide one useful next
+source. You do not need to prepare a Lab, repository or SSH connection first.
+See [first-use examples and checks](first-use-access.md). A web Project can hold
+uploaded sources; it does not automatically connect a laptop folder. Saving and
+live access depend on the tools available in the current chat.
+
+## Start with your question and sources
+
+Open a fresh chat with the Researcher and explicitly supply a checkout, source
+pack or permitted documentation. You do not need to classify the Surface.
+
+> Explain the AI-relevant components and paths in this source for a security
+> engineer. Save the Surface in this folder and maintain an investigation journal
+> as we go. Show me what is known, what remains uncertain and useful next steps.
+
+Expect a useful initial plan and a shallow map, then evidence-driven investigation.
+The Researcher uses this chat's supplied sources. It should not search old chats or
+vault research automatically. If the host supplies recognizable historical context,
+it should disclose that context; ask it to exclude it when inappropriate. This
+is not a promise that the host has no memory.
+
+## Start with an existing system
+
+If your question concerns a running composition, identify the instance and the
+read-only access you permit. Supply relevant remote repositories, documents or
+URLs through tools available in your Codex/ChatGPT host. A URL alone does not
+provide private access, and a local checkout may describe only one component.
+Candidate 0.1.15 follows this [instance-first workflow](instance-first-research.md).
+
+> Understand this running system using the host access I have provided. Explain
+> its components and connections, what is configured, what has been observed and
+> what remains untested. Keep inspection read-only and save the Surface here.
+
+Expect partial progress when an integration cannot be inspected. Researcher does
+not start a service, sign in for you or send a test prompt to complete the graph.
+On a later visit, it distinguishes actual target changes from newly learned facts
+and missing access. The existing source-only workflow remains available.
+
+## Inspect the developing Surface
+
+Open the latest brief for the current explanation, the HTML map for paths and
+expandable evidence, and the journal for the investigation journey. The Researcher
+should provide the working folder and absolute file links without changing projects.
+The journal retains scope, available evidence, findings, choices, questions,
+diversions, deferred paths and reasons. Each next step explains its purpose and
+needed access. You can inspect the whole map, redirect or stop at any time.
+
+A supported database is a possibility; it is not necessarily the selected target.
+Safe static configuration may establish selection, while a placeholder cannot.
+Missing configuration means unknown selection. Solid paths are selected; dashed
+paths require inspecting their support/selection details. Dated observed use is
+separate. Optional does not mean weak evidence.
+
+## Redirect and save
+
+> Pause this path and investigate authentication first. Keep the deferred work,
+> explain what access would help, and save the updated Surface.
+
+Expect the choice and its reason in the journal, useful next options and a new
+numbered revision. Previous files remain unchanged. Saving generates a brief,
+map and journal from the same Surface; there is no separate native data authority.
+The technical file is CycloneDX 1.7 with a bounded Surface profile. Ordinary users
+need not choose that format or ask for a separate export.
+
+## Return in a fresh chat
+
+Supply the latest `surface.cdx.json` and selected sources. The BOM alone retains
+relationships, evidence, choices, uncertainty and the investigation journey.
+
+> Continue this Surface. What did I choose, what remains uncertain, and what
+> should we investigate next? Keep earlier decisions and revisions.
+
+The Researcher reads the supplied Surface, re-establishes access and tells you
+when evidence cannot be refreshed. A new observation does not rewrite old ones.
+If you edit a generated journal, supply that edited folder and ask to reconcile
+it. The Researcher treats edits as proposals, records accepted changes in a new
+revision and preserves the edited original. It must not silently overwrite it.
+
+Old native/exchange saves remain readable when deliberately supplied. Migration
+requires reviewing their evidence and recording limits; no automatic lossless
+conversion is claimed. See the [format details](skill-cyclonedx.md).
+
+## What the fresh rehearsal showed
+
+The installed Researcher saved an initial plan before inspecting sources, then a
+second revision with the partial graph and next options. It worked from the
+explicitly supplied checkout, without the old operational notes. It reported that
+Git inspection was unavailable in its session and used inspected-file hashes.
+That is a useful distinction: source identity can be bounded without pretending
+the entire checkout or deployment has been verified.
+
+A scripted user diversion successfully moved the investigation to authentication.
+The journal kept the tool/document paths as deferred, with reasons, and retained
+the choice as a rehearsal decision. Earlier files were unchanged. The agent's
+presentation check initially missed a setting name because Markdown escaped its
+underscores; it normalized that escaping and completed readback. If a check fails,
+ask what failed and how readback was verified, rather than accepting a bare success
+claim.
+
+The fresh continuation received only a copied BOM and source access. It recovered
+the authentication-first choice and deferred work, proposed two next steps with
+access requirements, and saved another revision. Earlier files, relationships and
+source evidence remained intact. This exercises the ordinary “continue this
+Surface” interaction; human visual review and acceptance remain separate.
+
+## Report your check
+
+Pass when the brief is useful, evidence is inspectable, choices survive reopening
+and earlier files remain intact. Fail for invented relationships, lost decisions,
+missing state or unsupported claims of observed use. Blocked means a prerequisite
+such as source access or Node is unavailable. Return the interaction and relevant
+file paths without secrets. Agent rehearsals do not substitute for your acceptance.
+
+The Researcher describes and investigates; it does not deploy, attack, certify or
+score the target. A schema check is not evidence that a selected connection works.