knodin 0.7.6 → 0.8.2

package/docs/SIGNED-UPDATES.md

@@ -0,0 +1,151 @@
+# Signed update metadata trust
+
+knodin's update trust is local and channel-independent. npm, Homebrew,
+Artifactory, GitHub SaaS, and GHES may transport metadata and artifacts, but no
+one transport is an authority to install code.
+
+`src/update-trust.ts` is the pure verification boundary. It performs no network
+request, package-manager command, filesystem activation, or telemetry write.
+`src/update-policy.ts` composes that verifier with bounded HTTPS fetching,
+private monotonic state, policy, channel cross-checking, quarantine, exact local
+manager artifacts, health validation, and rollback. C64 adds the release-time
+CycloneDX/Sigstore evidence and fail-closed five-channel attestation machinery;
+C65 must still certify the smoke sandbox and recovery drills, and a real release
+must produce verified evidence before activation is enabled.
+
+## Trust model
+
+The metadata model has four disjoint Ed25519 roles:
+
+| Role | Purpose | Maximum signed lifetime | Required threshold |
+| --- | --- | ---: | ---: |
+| `root` | Delegates keys and rotates trust | 366 days | At least 2 |
+| `targets` | Authorizes immutable release artifacts and channels | 93 days | At least 2 |
+| `snapshot` | Binds an exact targets version, length, and digest | 32 days | At least 1 |
+| `timestamp` | Publishes the current snapshot and bounds freezes | 8 days | At least 1 |
+
+Root and targets thresholds prevent one stolen signing key from authorizing a
+release. All four roles must use separate, content-addressed keys. A root
+rotation is accepted one version at a time and only when both the previously
+trusted root threshold and the candidate root threshold sign the new root.
+
+The initial root envelope is trusted only when its SHA-256 matches an
+independently pinned digest shipped by a previously trusted knodin build. The
+pin must never be learned from fetched metadata or a package registry.
+
+Metadata is signed over deterministic UTF-8 JSON: object keys use ECMAScript
+code-unit ordering, arrays retain order, and undefined, cyclic, non-finite, or
+non-plain values are rejected. Key IDs are the SHA-256 of the canonical public
+key record.
+
+## Verification order
+
+The verifier fails closed in this order:
+
+1. Validate the installed trusted root and its independent digest pin.
+2. Verify the trusted root's own threshold signatures.
+3. Accept the same root byte-for-byte, or verify a one-step rotation under both
+   old and new root thresholds.
+4. Validate role separation, schema, generation time, expiry, and maximum
+   metadata lifetime.
+5. Verify threshold signatures for targets, snapshot, and timestamp.
+6. Bind timestamp to the exact canonical snapshot envelope by version, byte
+   length, and SHA-256.
+7. Bind snapshot to the exact canonical targets envelope the same way.
+8. Compare every role's version and digest with durable prior state. Lower
+   versions are rollback; changed bytes at the same version are equivocation.
+9. Select only the requested relative target path.
+10. Before quarantine, compare the downloaded artifact with its signed byte
+    length and SHA-256.
+
+The successful result returns the exact verified root envelope. C63 persists it
+for evidence, but never promotes a root merely because mutable local state says
+it is trusted: each check and rollback remains anchored to the independently
+shipped initial root and pin. When more than one rotation was missed, the client
+fetches each numbered intermediate root, bounds one check to 32 rotations,
+cross-checks every intermediate across configured mirrors, and advances only
+one dual-threshold version at a time. A skipped, substituted, malformed, or
+wrong-version intermediate root fails closed without persisting new trust.
+
+The signed target also carries the semantic version, release channel, full
+source commit, and digests for build provenance, the CycloneDX SBOM, its signed
+attestation, and the cross-channel release attestation. Metadata validation
+accepts these fields so notify-only clients can read a staged repository, but
+download/apply refuses when any digest is absent. The verifier does not invent
+missing provenance or promote a merely claimed channel result.
+
+## Defenses and limits
+
+Executable tests cover valid threshold authorization, missing threshold
+signatures, signed-field mutation, artifact substitution, expired timestamp
+freeze, rollback, same-version equivocation, mix-and-match metadata, role-key
+reuse, root-pin substitution, future metadata, and root rotation without the
+old threshold. Recovery tests additionally cover missed sequential rotations,
+skipped roots, intermediate mirror disagreement, and malicious version jumps.
+
+An isolated 0.4.3 adversarial drill executed the trust, policy, CLI, and release-
+attestation suites together: 4 files and 52 scenarios passed on macOS with Node
+26. Its machine-readable record is
+`docs/evidence/0.4.3-update-defense-drill.json`. The record separately identifies
+the post-tag commit from which the drill executed and the immutable 0.4.3 source
+commit whose implementation it exercised, plus the environment, command,
+defenses, and exclusions. Injected transport, package-manager, smoke, and health
+adapters make this repeatable evidence of the verification and recovery logic;
+they do not certify a real npm/Homebrew mutation path or replace the production
+ceremony and live drills.
+
+The verifier cannot make the first installation trustworthy by itself. A
+production release still requires an offline root-key ceremony, independent
+pin distribution, durable trusted-state storage, compromised-key rotation
+drills, and verified artifacts on every approved channel. Until that ceremony
+and C64–C65 are complete:
+
+- the former npm `latest` hint is not used by doctor, status, MCP, or update commands;
+- no release is authorized by `src/update-trust.ts` in production;
+- no automatic update may be enabled; and
+- a registry, formula, release page, or mutable URL must not be treated as a
+  root of trust.
+
+## Production key ceremony gate
+
+Private root keys must be created and retained offline under separate
+custodians. No production private key or recovery secret belongs in this
+repository, CI variables, npm, Homebrew, GitHub, GHES, or Artifactory. Before
+activating signed checks, the release owners must record:
+
+- custodians and a two-person approval path;
+- public key IDs and the independently reviewed root-envelope digest;
+- offline backup and recovery controls;
+- online targets/snapshot/timestamp key storage and rotation periods;
+- emergency revocation and root-rotation rehearsal evidence; and
+- the exact knodin source commit that first embeds the trusted root pin.
+
+This external ceremony is deliberately a release gate rather than generated
+test material masquerading as production trust.
+
+The fail-closed preparation contracts and multi-custodian rehearsal procedure
+are in [`ROOT-CEREMONY-RUNBOOK.md`](ROOT-CEREMONY-RUNBOOK.md). Their import seam
+accepts public material only and always returns `productionActivation: false`;
+it does not alter the production `trust-unconfigured` behavior.
+
+## Client policy and recovery
+
+The implemented policy modes are `notify-only`, `download-verify-only`,
+`apply-patch`, and `apply-minor`; major updates always require manual approval.
+Policies also bound release age, approved channels, enterprise mirror origins,
+optional exact metadata/artifact cross-checks, and UTC maintenance windows.
+
+`knodin update status` and `explain` are offline reads. `check` uses no
+repository context. `apply` first quarantines and verifies the candidate plus a
+signed artifact for the currently installed version. A missing rollback target,
+provenance object, SBOM, release attestation, smoke sandbox, manager adapter, or
+health check is a refusal, not a warning. Explicit rollback re-verifies the
+stored signed metadata, artifact, provenance, SBOM, both attestations, and every
+quarantine path before invoking the manager; nested symlink redirection is a
+hard refusal. Apply also requires the candidate attestation's rollback version,
+path, and derived digest to match the separately downloaded and verified rollback
+artifact. Signed
+non-SemVer target versions are ineligible before path construction; quarantine
+uses fresh exclusive files under a symlink-rejecting private root; and automatic
+rollback succeeds only after a second health check validates the restored
+installation.

package/docs/TELEMETRY.md

@@ -0,0 +1,46 @@
+# Local savings telemetry
+
+knodin's optional telemetry is local, metadata-only, and disabled by default.
+It uses `gpt-tokenizer@3.4.0` with the `o200k_base` encoding and records:
+
+- the actual optimized response contract, bytes, and tokens;
+- an executed full-file baseline only when the operation identifies bounded
+  repository-contained files;
+- byte and token savings, including negative savings;
+- latency p50/p95 inputs, process RSS, cold/warm operation state, success/error
+  outcome, agent round trips, truncation, detail mode, and MCP schema token cost;
+- freshness failures, compression-fidelity samples, indexing time, and current
+  local index storage where those measurements are available;
+- a one-way SHA-256-derived repository identifier for per-repository grouping.
+
+Unknown baselines remain `null`; knodin does not invent a full-file comparison.
+Persisted records contain no source or file paths. Enable persistence for an MCP
+call with `persistTelemetry: true`. Records default to 30-day retention and are
+stored mode `0600` with atomic replacement. Inspect and manage them with:
+
+```text
+knodin telemetry status
+knodin telemetry report
+knodin telemetry export
+knodin telemetry clear
+```
+
+`--retention-days 1..3650` changes the read/persistence window. `--input` and
+`--output` accept only repository-contained, non-symlink paths. `clear` is the
+only command that deletes the private JSONL file; report and export never do.
+
+The default report is `.knodin/telemetry-report.html`. It includes summary,
+operation, time-series, private repository, indexing-cost, latency, freshness,
+and fidelity views with light/dark styling. The default machine-readable export
+is `.knodin/telemetry-export.json` and labels measured, modeled, and unknown
+baselines explicitly. Neither artifact needs a daemon, hosted service,
+authentication, or source egress.
+
+Telemetry never persists source, raw tool output, command text, usernames, or
+absolute paths. A measured baseline is an executed, bounded, repository-local
+full-file read; it is not inferred from file counts. Unsupported baselines and
+metrics remain `null` or are counted as unavailable instead of being modeled.
+
+ROI telemetry is separate from troubleshooting diagnostics. Telemetry measures
+product behavior; `knodin diagnostics` retains sanitized failure envelopes and
+creates a user-inspectable support bundle. See `docs/DIAGNOSTICS.md`.

package/docs/TOKEN-OPTIMIZER-SCORECARD.md

@@ -0,0 +1,79 @@
+# Reckon vs Token Optimizer — evidence-linked scorecard
+
+This scorecard evaluates Token Optimizer at immutable commit
+`a2b9cef5efdfcf0393e9e7a5ded229ffd9f613d7`. The machine-auditable
+[scorecard manifest](../benchmarks/evaluations/token-optimizer-20260730/scorecard.json)
+links every disposition to implementation, verification source/test paths, benchmark
+evidence, and a limitation. The [methodology](../benchmarks/evaluations/token-optimizer-20260730/methodology.md)
+and [raw replay](../benchmarks/evaluations/token-optimizer-20260730/raw-behavior-replay-20260730.json)
+are the authority for measured values.
+
+## Executive result
+
+Reckon passes the pinned operation-level oracle for all five safe structural
+workflows and is smaller in real tokens and faster at warm p50/p95 for each. The
+oracles are deliberately narrow: three Python names, one byte-exact Python
+method body, two exact file sets, and one eligible-file count. They do not prove
+signature, nesting, kind, or construct accuracy across all seven languages.
+Reckon has the stronger
+verified compression budget/recovery contract, local opt-in real-token
+telemetry, graph-backed failure diagnosis, freshness evidence, lifecycle
+automation, and implemented signed-update machinery. These are bounded results,
+not proof that Reckon is better in every repository or environment.
+
+Knodin does not copy the competitor's unrestricted shell command runner. C58's
+portable prototype remains rejected. C88 instead ships immutable, doubly
+enabled repository profiles on the certified macOS boundary; Linux and Windows
+fail closed as `containment-unavailable` pending native adapters.
+Reckon's onboarding readiness remains incomplete because native Windows
+named-client and timed two-minute trials are missing. Trusted distribution is implemented but
+not production-complete: the root ceremony, five-channel signed release,
+sandbox/manager drills, and adversarial live exercises remain C62–C67 gates.
+
+| Dimension | Comparative disposition | Readiness | Reproducible result and boundary |
+|---|---|---|---|
+| File outline | Stronger on pinned oracle | Implemented/verified | Three Python names pass; 56 vs 64 tokens and 0.203 vs 0.478 ms warm p50; cold startup is slower |
+| Exact symbol source | Stronger on pinned oracle | Implemented/verified | One Python body is byte-exact; 33 vs 37 tokens and 0.258 vs 0.419 ms warm p50 |
+| Symbol search | Stronger on pinned oracle | Implemented/verified | Exact seven-file set; 76 vs 84 tokens and 0.227 vs 1.112 ms warm p50 |
+| Batch outline | Stronger on pinned oracle | Implemented/verified | Exact seven-file set; 331 vs 377 tokens and 0.184 vs 1.387 ms warm p50; Reckon bytes are larger |
+| Project overview | Stronger on pinned oracle | Implemented/verified | Exact eligible-file count; 14 vs 127 tokens and 0.135 vs 0.589 ms warm p50 |
+| Parser/language fidelity | Not comparatively verified | Implemented with limitations | Seven languages participate, but no shared construct-accuracy oracle exists |
+| Output compression | Stronger on pinned oracle | Implemented/verified | 7/7 signals in 14/20 lines with hard recovery; competitor is faster warm, and Reckon's full envelope is 18 tokens larger |
+| Command execution | Stronger control contract; portability limited | Implemented on certified macOS boundary | No shell/caller argv, double opt-in, sanitized environment, network/user-data/fork denial, hard time/output bounds, compression + diagnosis; Linux/Windows fail closed |
+| Telemetry/privacy | Stronger source contract | Implemented/verified | Reckon is opt-in with real tokens; competitor defaults on and records sensitive paths/command text |
+| Dashboard UX | Not comparatively verified | Implemented/verified | Both have local HTML; only Reckon's rendering/data contract is exercised here |
+| Structural depth/freshness | Additional capability | Implemented/verified | Stable identity, relationships, and commit freshness; no comparative production study |
+| Failure-to-code intelligence | Additional capability | Implemented/verified | Retained diagnostics map to code/tests/impact; static relations are not runtime proof |
+| Cross-platform onboarding | Not comparatively verified | Implemented with limitations | macOS/Linux gates exist; timed and native Windows named-client trials are absent |
+| Lifecycle/portfolio | Additional capability | Implemented with limitations | Live 61-repository bounded certification; async refresh can still require repair |
+| Installation/maintenance | Stronger source contract | Implemented with limitations | Packaged install/uninstall versus competitor clone/venv/manual registration; Windows certification remains open |
+| Trusted distribution | Stronger source contract | Production incomplete | Competitor has no trust mechanism; Reckon ceremony, drills, and five-channel release remain open |
+| Natural routing | Claim not reproducible | Unverified claim | Neither product has a controlled versioned routing-frequency dataset |
+| Claimed savings | Claim not reproducible | Unverified claim | Competitor percentages lack equivalent-case raw production data |
+
+## What is factual to say
+
+It is factual to say that Reckon is a more complete local code-intelligence
+system than the pinned Token Optimizer implementation: it passes the pinned
+operation-level structural and compression oracles and adds graph
+relationships, impact/review context, failure diagnosis, freshness, lifecycle,
+portfolio, private measurement, and update-trust machinery.
+
+It is not yet factual to say “better across all dimensions” or “production
+supply-chain complete.” Contained execution is macOS-only, onboarding
+certification has named gaps, and the production trust
+and distribution exercises remain open. Release-facing language must retain
+those qualifications until C60, C62–C67 close with live evidence.
+
+## Reproduce
+
+```bash
+npm run audit:token-optimizer
+REPLAY_OUTPUT="$(mktemp -t reckon-token-optimizer-replay.XXXXXX.json)"
+npm run bench:token-optimizer -- "$REPLAY_OUTPUT"
+npx vitest run src/__tests__/unit/docs-integrity.spec.ts
+```
+
+The source audit requires authenticated GHES access. The replay is pinned to
+immutable competitor blobs and records cold/warm latency, serialized bytes,
+real `o200k_base` tokens, shared correctness oracles, and explicit gaps.

package/docs/assets/knodin-favicon.svg

@@ -0,0 +1,4 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 64 64" width="64" height="64">
+  <path d="M46,6 L4,32 L46,58 L46,44 L26,32 L46,20 Z" fill="#130032" />
+  <path d="M58,24 L58,40 L34,32 Z" fill="#4C00FF" />
+</svg>

package/docs/releases/0.3.0.md

@@ -0,0 +1,46 @@
+# Reckon Graph 0.3.0 release notes
+
+Reckon Graph 0.3.0 makes graph freshness revision-truthful and closes the
+lifecycle and completion-audit gaps observed in a large multi-worktree field
+run. It is a backward-compatible minor release.
+
+## Freshness and lifecycle
+
+- Every graph answer carries a canonical current/indexed revision and
+  dirty-working-tree freshness envelope.
+- Durable, coalescing hook events survive contention and failures; status,
+  `wait --fresh`, and `status --watch` expose queue progress.
+- Deep repair binds a revision only after complete content verification.
+- Worktree status, reconciliation, and safe dry-run removal expose per-worktree
+  graph ownership and detect removed worktrees.
+
+## Diagnostics and completion audits
+
+- Doctor reports separate Claude, Codex, Gemini, and Antigravity configuration,
+  protocol checks, and active-session evidence.
+- Repair plans distinguish health findings from planned operations.
+- PR listing and audits return explicit repository/login/error/zero-result
+  context, reviews, threads, checks, merge commits, branches, and graph impact.
+
+## Bounded structural context
+
+- File and batch outlines include stable identity, signatures, ranges,
+  visibility, containment, and parser evidence.
+- Source-only explain and deterministic project overview reduce unnecessary
+  full-file reads.
+- Go and Rust outlines use checked-in local Tree-sitter grammars.
+- Optional local telemetry uses
+  `gpt-tokenizer@3.4.0:o200k_base`, preserves negative savings, measures schema
+  cost, and can render a static repository-local report without egress.
+
+Command execution/output compression was evaluated but is not shipped because
+portable hard memory and process-tree containment were not proven. Competitor
+behavioral measurements remain incomplete without an authorized packaged
+release or raw usage dataset. A post-release audit now pins and verifies the
+authorized source revision without treating source inspection as end-to-end
+performance evidence. See
+[`../RELEASE-0.3-EVIDENCE.md`](../RELEASE-0.3-EVIDENCE.md) and
+[`../COMMAND-OUTPUT-COMPRESSION.md`](../COMMAND-OUTPUT-COMPRESSION.md).
+
+No tag, npm publication, Homebrew publication, or P&T mirror is created by this
+release PR; each remains a separately authorized release action.

package/docs/releases/0.4.0.md

@@ -0,0 +1,68 @@
+# Reckon Graph 0.4.0 release notes
+
+Reckon Graph 0.4.0 delivers the implemented portions of the Token Optimizer
+parity program and adds trusted-update, release-attestation, large-portfolio,
+and compact-response capabilities. It is a backward-compatible minor release.
+
+## Compact structural and diagnostic context
+
+- Compact file and batch outlines, symbol extraction, symbol search, and
+  project overview reduce response tokens while retaining stable identity,
+  source locations, parser evidence, ambiguity, and freshness.
+- Deterministic log compression preserves diagnostic structures, enforces hard
+  budgets, records omissions, and supports bounded retrieval without rerunning
+  the original command.
+- Failure diagnosis links log evidence to files, symbols, callers,
+  dependencies, tests, and recent changes in the local graph.
+- The checked-in Token Optimizer replay and evidence-linked scorecard pin the
+  competitor and Reckon revisions and keep unsupported claims explicit.
+
+## Portfolio and lifecycle reliability
+
+- Repository filters are applied before inventory, large `git ls-files`
+  responses are supported, and selected repositories are processed
+  sequentially.
+- Portfolio initialization is bounded, resumable, and truthful in dry-run
+  mode, including nested repositories and noisy Salesforce metadata.
+- Lifecycle queues, `wait --fresh`, and installed-versus-active client status
+  provide revision-truthful observability.
+- Terminal help uses a declarative command model and the available terminal
+  width while remaining snapshot-testable.
+
+## Private ROI evidence
+
+- Local, opt-in telemetry uses a real tokenizer and records bounded operation,
+  latency, indexing, schema, fidelity, and savings metrics without source,
+  command text, absolute paths, usernames, or network egress.
+- The self-contained dashboard and export bundle distinguish directly measured
+  observations from modeled counterfactuals.
+
+## Trusted distribution and updates
+
+- Signed-only update status, check, explain, apply, and rollback policies bind
+  artifact length/digest, channel, provenance, SBOM, release age, maintenance
+  window, and rollback state.
+- TUF-inspired root, targets, snapshot, and timestamp verification detects
+  rollback, freeze, mix-and-match, expiry, threshold, and compromised-channel
+  failures. Bounded sequential root recovery handles missed rotations.
+- The release workflow builds one tarball, creates and independently verifies
+  provenance and CycloneDX SBOM attestations, then publishes the exact bytes to
+  npm and Docusign Artifactory.
+- Final certification additionally requires byte-identical GitHub SaaS, GHES,
+  and Homebrew observations plus the machine-readable cross-channel release
+  attestation.
+
+## Intentional boundary
+
+Reckon does not expose unrestricted shell execution. Portable process-tree,
+filesystem, network, credential, and resource containment has not been proven,
+so command execution remains an explicit safety-based non-goal. Existing
+trusted execution tools can pass their output to Reckon's compression and
+failure-diagnosis operations.
+
+Version preparation does not itself publish any channel. The release is
+certified only after the separately authorized tag workflow and five-channel
+verification complete. Production offline root custody, compromised-runner
+drills, automated non-personal GHES mirroring, timed onboarding, and native
+Windows named-client certification remain documented limitations until their
+evidence is recorded.

package/docs/releases/0.4.1.md

@@ -0,0 +1,28 @@
+# Reckon Graph 0.4.1 release notes
+
+Reckon Graph 0.4.1 is the publishable patch release of the 0.4 line. It contains
+all functionality described in the 0.4.0 release notes plus a fail-closed
+release-verification compatibility fix.
+
+## Release verification fix
+
+- The workflow now verifies signed provenance and the CycloneDX SBOM using the
+  exact certificate identity together with repository, source digest, source
+  ref, predicate type, and GitHub Actions OIDC issuer constraints accepted by
+  the current GitHub CLI.
+- The release-attestation CLI test double now runs with `set -eu`, ensuring an
+  assertion mismatch actually fails the test instead of continuing to a later
+  successful command.
+
+## Why 0.4.0 was not published
+
+The immutable `v0.4.0` tag reached the signed-evidence stage, then failed closed
+before npm or Artifactory publication because the installed GitHub CLI rejects
+combining `--cert-identity` with `--signer-workflow`. The tag remains unchanged
+for auditability and 0.4.0 was not published to any package channel. This patch
+release contains the corrected workflow rather than rewriting the existing
+tag.
+
+As with every release, version preparation alone does not certify distribution.
+Certification requires the tag workflow and byte-for-byte verification across
+npm, Docusign Artifactory, GitHub SaaS, GHES, and Homebrew.

package/docs/releases/0.4.2.md

@@ -0,0 +1,27 @@
+# Reckon Graph 0.4.2 release notes
+
+Reckon Graph 0.4.2 is the publishable patch release of the 0.4 line. It contains
+all functionality described in the 0.4.0 and 0.4.1 release notes plus a
+canonical CycloneDX attestation-predicate fix.
+
+## CycloneDX attestation verification fix
+
+- The release workflow and final cross-channel verifier now use the canonical
+  `https://cyclonedx.org/bom` predicate type emitted by GitHub's SBOM
+  attestation action. The CycloneDX document's `specVersion` remains validated
+  independently as 1.4, 1.5, or 1.6.
+- Regression tests require the canonical attestation predicate for SBOM bundles
+  and reject accidental coupling to a document-version URI.
+
+## Why 0.4.1 was not published
+
+The immutable `v0.4.1` tag produced a verified release tarball, signed SLSA
+provenance, and a signed CycloneDX SBOM, then failed closed before npm or
+Artifactory publication. Inspection of the signed bundle proved that GitHub
+emitted `https://cyclonedx.org/bom`, while the verifier requested
+`https://cyclonedx.org/bom/v1.5`. The tag remains unchanged for auditability;
+this patch release corrects the verifier instead of rewriting it.
+
+Version preparation alone does not certify distribution. Certification still
+requires the tag workflow and byte-for-byte verification across npm, Docusign
+Artifactory, GitHub SaaS, GHES, and Homebrew.

package/docs/releases/0.4.3.md

@@ -0,0 +1,23 @@
+# Reckon Graph 0.4.3
+
+Reckon Graph 0.4.3 corrects the approved Docusign Artifactory channel identity
+used by final release attestation.
+
+## Release-integrity correction
+
+- Final channel evidence now requires the authoritative
+  `artifactory-corp.docusignhq.com` host returned by Docusign Artifactory.
+- The obsolete `artifactory.docusignhq.com` alias is rejected. It does not
+  serve the published raw artifact and must not appear in a verified release
+  record.
+- Regression fixtures cover the production host and prevent the obsolete alias
+  from returning.
+
+Version 0.4.2 was successfully published to npm, Artifactory, GitHub SaaS, and
+Homebrew, but its final five-channel attestation was intentionally withheld when
+the real Artifactory observation exposed this host mismatch. No mismatched
+artifact was installed: all four observed 0.4.2 artifacts had identical bytes.
+
+As with every release, 0.4.3 is certified only after the tag workflow and final
+cross-channel verification complete successfully. Preparing this version does
+not itself publish it.

package/docs/releases/0.5.0.md

@@ -0,0 +1,29 @@
+# knodin 0.5.0
+
+knodin 0.5.0 completes the public product and package rename from Reckon Graph.
+
+## New identity
+
+- The unscoped npm package is `knodin`.
+- The canonical executable is `knodin`.
+- The MCP server and its single gateway tool are named `knodin`.
+- The display name is always lowercase, with the caret-beak mark as the
+  canonical logo and `<◀  knodin` as the terminal lockup.
+
+## Compatibility
+
+- The package still exposes `reckon` as a compatibility executable.
+- Existing `.reckon/` state, `reckon.yaml`, and `RECKON_*` environment
+  variables are retained so the rename does not force a reindex or invalidate
+  local configuration.
+- `knodin init` migrates managed MCP registrations and recognizes lifecycle
+  hooks created by Reckon Graph.
+- Historical benchmark and release evidence retains its original product name
+  and source identities.
+
+## Distribution
+
+Public npm is the initial supported installation channel. The previous
+`reckon-graph` package remains available as a deprecated compatibility
+signpost; it is not unpublished. The legacy Homebrew formula is not a supported
+knodin 0.5.0 channel.

package/docs/releases/0.5.1.md

@@ -0,0 +1,17 @@
+# knodin 0.5.1
+
+knodin 0.5.1 makes the lowercase product name canonical across every public
+distribution surface.
+
+## Canonical identity
+
+- The product and npm package use the `knodin` name.
+- Correction: `knodin/knodin` is the Homebrew-tap repository. The authoritative
+  source repository remains `DTS-Productivity-Engineering/reckon-graph` until a
+  separately authorized GitHub repository rename occurs.
+
+## npm branding
+
+- The caret-beak SVG is included in the published package.
+- The README loads the logo through the public npm CDN, so it renders on npm
+  even while the source repository remains private.

package/docs/releases/0.6.0.md

@@ -0,0 +1,18 @@
+# knodin 0.6.0
+
+This release completes the operational product rename.
+
+- Repository-local graph, hook, lifecycle, telemetry, and generated state now
+  lives under `.knodin/`.
+- Team system configuration is now `knodin.yaml`.
+- Runtime environment variables use the `KNODIN_` prefix.
+- Managed Git hooks, agent instructions, MCP registrations, and configuration
+  markers use only `knodin` identifiers.
+- The package installs only the `knodin` executable; the former command alias
+  has been removed.
+- The npm README is reorganized around product value, measured competitive
+  results, installation, first use, and capability reference.
+
+Existing local graph directories must be renamed from the former hidden-state
+path to `.knodin` before using this release. Re-run `knodin init` afterward to
+refresh managed hooks and agent configuration.

package/docs/releases/0.7.0.md

@@ -0,0 +1,24 @@
+# knodin 0.7.0
+
+This release is the dogfood-readiness release for local code intelligence with
+bounded, user-controlled troubleshooting evidence.
+
+- Adds `knodin diagnostics enable|status|collect|inspect|clear|disable` with a
+  private, bounded local failure journal and redacted support bundles. Nothing
+  is uploaded automatically.
+- Captures sanitized CLI and MCP failure correlation IDs without retaining
+  source, queries, arguments, environment values, credentials, or raw paths.
+- Adds contained execution profiles with explicit platform support and
+  fail-closed behavior when the requested containment cannot be established.
+- Adds structural cold-start routing and checked-in C87 latency evidence while
+  preserving truthful response budgets and freshness metadata.
+- Integrates lifecycle refresh through hook-manager-native Lefthook routing,
+  while preserving repository-owned configuration and distinguishing graph
+  repair from lifecycle reinitialization.
+- Tightens freshness proof, response-budget accounting, package smoke coverage,
+  and roadmap evidence for the completed local feature set.
+
+This ordinary 0.x release is intended for normal installation, including the
+initial Docusign rollout. The external evidence and signed-release gates in
+`roadmap/competitive-roadmap.md` remain open and continue to block GA and
+trusted-distribution claims; publication does not waive or simulate them.

package/docs/releases/0.7.1.md

@@ -0,0 +1,21 @@
+# knodin 0.7.1
+
+This is the first normally published 0.7 release of knodin. It provides local,
+opt-in diagnostics that users can collect and deliberately share when they
+need troubleshooting help; it is not a special dogfood build or prerelease.
+
+- Adds `knodin diagnostics enable|status|collect|inspect|clear|disable` with a
+  private, bounded local failure journal and redacted support bundles. Nothing
+  is uploaded automatically.
+- Captures sanitized CLI and MCP failure correlation IDs without retaining
+  source, queries, arguments, environment values, credentials, or raw paths.
+- Adds contained execution profiles, structural cold-start routing, lifecycle
+  refresh improvements, tighter freshness proof, and truthful response-budget
+  accounting.
+- Corrects the release test suite so the competitive sandbox contract is
+  verified on systems both with and without Bun installed.
+
+This ordinary 0.x release is suitable for installation through npm,
+Artifactory, and Homebrew, including use by Docusign employees. The external
+evidence and signed-release gates recorded in `roadmap/competitive-roadmap.md`
+remain open and continue to block GA and trusted-distribution claims.

package/docs/releases/0.7.2.md

@@ -0,0 +1,21 @@
+# knodin 0.7.2
+
+This is the first normally published 0.7 release of knodin. It provides local,
+opt-in diagnostics that users can collect and deliberately share when they
+need troubleshooting help; it is not a special dogfood build or prerelease.
+
+- Adds `knodin diagnostics enable|status|collect|inspect|clear|disable` with a
+  private, bounded local failure journal and redacted support bundles. Nothing
+  is uploaded automatically.
+- Captures sanitized CLI and MCP failure correlation IDs without retaining
+  source, queries, arguments, environment values, credentials, or raw paths.
+- Adds contained execution profiles, structural cold-start routing, lifecycle
+  refresh improvements, tighter freshness proof, and truthful response-budget
+  accounting.
+- Makes release verification deterministic across macOS and Linux, both with
+  and without Bun, including foreground execution of lifecycle-hook fixtures.
+
+This ordinary 0.x release is suitable for installation through npm,
+Artifactory, and Homebrew, including use by Docusign employees. The external
+evidence and signed-release gates recorded in `roadmap/competitive-roadmap.md`
+remain open and continue to block GA and trusted-distribution claims.

package/docs/releases/0.7.3.md

@@ -0,0 +1,23 @@
+# knodin 0.7.3
+
+This is the first normally published 0.7 release of knodin. It provides local,
+opt-in diagnostics that users can collect and deliberately share when they
+need troubleshooting help; it is not a special dogfood build or prerelease.
+
+- Adds `knodin diagnostics enable|status|collect|inspect|clear|disable` with a
+  private, bounded local failure journal and redacted support bundles. Nothing
+  is uploaded automatically.
+- Captures sanitized CLI and MCP failure correlation IDs without retaining
+  source, queries, arguments, environment values, credentials, or raw paths.
+- Adds contained execution profiles, structural cold-start routing, lifecycle
+  refresh improvements, tighter freshness proof, and truthful response-budget
+  accounting.
+- Makes release verification deterministic across macOS and Linux, both with
+  and without Bun, including foreground execution of lifecycle-hook fixtures.
+- Ensures `configure --scope cli-only` removes repository-detected Knodin MCP
+  configuration even when that agent's CLI is not installed locally.
+
+This ordinary 0.x release is suitable for installation through npm,
+Artifactory, and Homebrew, including use by Docusign employees. The external
+evidence and signed-release gates recorded in `roadmap/competitive-roadmap.md`
+remain open and continue to block GA and trusted-distribution claims.