[gobby-cli-#1012] chore: prepare 2026-07-01 release

Author: joshwilhelmi

CHANGELOG.md

@@ -9,6 +9,83 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.4.0] — gcode — 2026-07-01
+
+### Added
+
+#### gcode
+
+- **Agentic CodeWiki aggregate generation** — aggregate CodeWiki pages can route
+  through gcore's daemon-side agentic/tool-backed generation path, using
+  read-only gcode investigation tools and recording lane/tool-call observability
+  in page frontmatter.
+- **CodeWiki purge** — `gcode codewiki --purge --out <DIR> --force` removes
+  generated CodeWiki Markdown and metadata under the selected output directory
+  without running generation or AI calls.
+- **Richer handbook output** — CodeWiki now emits a broader handbook/navigation
+  layer with generalized taxonomy, semantic clusters, reference tables,
+  per-category render versions, and bounded dependency diagrams on repo
+  overviews.
+- **Verification scope control** — `--ai-verify-scope aggregates|all` keeps the
+  default verifier pass on aggregate/curated pages while allowing slower
+  per-file verification when explicitly requested.
+
+### Fixed
+
+#### gcode
+
+- **Strict generated Markdown normalization** — CodeWiki output is normalized
+  before write/reuse so generated pages remain stable, markdownlint-friendly,
+  and idempotent across reruns.
+- **CodeWiki filtering and recovery** — `docs/` and temporary directories are
+  excluded from core-file generation; Lane B curated generation now investigates
+  reliably, bounds seed prompts, exposes AI HTTP error bodies, and hard-fails
+  ungroundable aggregate output instead of silently emitting skeleton prose.
+- **AI metadata and timeout hardening** — nested AI frontmatter is preserved,
+  daemon profile/routing metadata is aligned, non-retryable timeouts are handled
+  correctly, and text generation timeout is raised for long aggregate pages.
+- **Index maintenance fixes** — retry timestamp columns, UTF-8 signature
+  truncation, and prune discovery-row preservation are corrected.
+
+## [0.7.0] — gobby-core — 2026-07-01
+
+### Added
+
+#### gobby-core
+
+- **Lane A/B generation foundation** — shared tier profiles, direct chat
+  transport, provider-neutral tool-loop generation, daemon chat transport, and
+  `tool_chat` capability support now live in `gobby-core` for gcode and gwiki
+  consumers.
+- **Daemon-side agentic chat** — `daemon_agentic_chat` posts messages,
+  project context, reasoning settings, max turns, and a serialized
+  `ToolPolicy` to the daemon route for agentic narrative generation.
+- **Config secret decryption support** — datastore-backed AI/config sources can
+  resolve `$secret:` values through gcore's envelope secret handling while
+  env-only sources continue to reject secret references explicitly.
+
+### Fixed
+
+#### gobby-core
+
+- **Tool-loop resilience** — generation parsing and stop-reason handling are
+  hardened so malformed or incomplete tool-loop output returns typed failures
+  instead of ambiguous empty content.
+
+## [0.7.0] — ghook — 2026-07-01
+
+### Added
+
+#### ghook
+
+- **Shared daemon URL resolution** — live dispatch, planned-shutdown preflight,
+  statusline POSTs, and diagnose output now use `gobby_core::daemon_url`, so
+  `GOBBY_DAEMON_URL`, `GOBBY_PORT`, and bootstrap `daemon_url` overrides are
+  honored consistently.
+- **Machine identity stamping** — hook envelopes now include local
+  `machine_id` and `os` when available, or a `machine_id_error` code when the
+  local machine identity file is missing or empty.
+
 ### Removed
 
 #### ghook
@@ -19,6 +96,29 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   or POST, so old hook configs fail quiet instead of falling through to the
   unknown-CLI Claude fallback.
 
+## [0.7.0] — gwiki — 2026-07-01
+
+### Added
+
+#### gwiki
+
+- **Scoped purge command** — `gwiki --project <ROOT> purge --yes` and
+  `gwiki --topic <NAME> purge --yes` delete generated/indexed wiki state for
+  the explicit scope across PostgreSQL, Qdrant, and FalkorDB. Global purge is
+  rejected.
+- **Shared generation tiers** — `gwiki ask --llm` and `gwiki compile` now route
+  generation through gcore tier profiles and the shared Lane B tool-loop
+  utilities.
+
+### Fixed
+
+#### gwiki
+
+- **AI routing metadata** — generation output preserves aligned AI routing and
+  frontmatter metadata across direct and daemon-backed paths.
+
+### Removed
+
 #### gwiki
 
 - **Raw Gemini transcript fallback** — `gwiki --raw` no longer parses Gemini

Cargo.lock

@@ -27,10 +27,10 @@ This workspace ships three Gobby CLI tools plus a shared library.
 
 | Crate | Binary | Version | Release tag |
 |---|---|---:|---|
-| `gobby-code` | `gcode` | `1.3.3` | `gcode-v1.3.3` |
-| `gobby-core` | n/a | `0.6.1` | `gobby-core-v0.6.1` |
-| `gobby-hooks` | `ghook` | `0.6.2` | `ghook-v0.6.2` |
-| `gobby-wiki` | `gwiki` | `0.6.5` | `gwiki-v0.6.5` |
+| `gobby-code` | `gcode` | `1.4.0` | `gcode-v1.4.0` |
+| `gobby-core` | n/a | `0.7.0` | `gobby-core-v0.7.0` |
+| `gobby-hooks` | `ghook` | `0.7.0` | `ghook-v0.7.0` |
+| `gobby-wiki` | `gwiki` | `0.7.0` | `gwiki-v0.7.0` |
 
 ### gcode — Code Search & Navigation
 
@@ -54,7 +54,10 @@ daemon stale-project graph cleanup without cwd project resolution.
 pages) from the index, with optional AI prose tiers (`--ai-depth`), validated
 Mermaid architecture diagrams (workspace topology plus runtime/AI-generation
 sequence diagrams), and citation-checked grounding against indexed source
-spans. gwiki indexes the output as normal vault documents.
+spans. Aggregate pages can use the daemon-side agentic/tool-backed generation
+route, and `gcode codewiki --purge --out <vault> --force` clears generated
+CodeWiki output before a clean rebuild. gwiki indexes the output as normal
+vault documents.
 
 For non-Gobby-managed projects, `gcode init` installs the bundled `gcode` skill
 for Claude Code, Codex, Droid, Grok, Qwen, and Antigravity CLI. Gobby-managed

README.md

@@ -1,6 +1,6 @@
 [package]
 name = "gobby-code"
-version = "1.3.13"
+version = "1.4.0"
 edition = "2024"
 rust-version = "1.88"
 authors = ["Josh Wilhelmi <hello@gobby.ai>"]
@@ -22,7 +22,7 @@ path = "src/main.rs"
 
 [dependencies]
 # Internal
-gobby-core = { path = "../gcore", version = "0.6.1", features = ["postgres", "falkor", "qdrant", "search", "indexing", "graph-analytics", "ai"] }
+gobby-core = { path = "../gcore", version = "0.7.0", features = ["postgres", "falkor", "qdrant", "search", "indexing", "graph-analytics", "ai"] }
 
 # CLI
 clap = { version = "4", features = ["derive"] }
@@ -96,7 +96,7 @@ tempfile = "3"
 libc = "0.2"
 
 [build-dependencies]
-gobby-core = { path = "../gcore", version = "0.6.1" }
+gobby-core = { path = "../gcore", version = "0.7.0" }
 
 [dev-dependencies]
 insta = { version = "1", features = ["json"] }

crates/gcode/Cargo.toml

@@ -81,6 +81,16 @@ for the UI, but graph sync/read/lifecycle behavior lives in `gcode`.
 - `gcode vector cleanup-orphans` — remove Qdrant code-symbol vectors for files missing from PostgreSQL, without resolving embeddings
 - `gcode prune` — remove stale project records globally and reconcile graph and vector projections for all remaining indexed projects; use `--project` to scope projection cleanup
 
+## CodeWiki Lifecycle
+
+- `gcode codewiki --out <vault>` — generate vault-ready code documentation under the selected output directory
+- `gcode codewiki --scope <PATH...> --out <vault>` — regenerate only docs for indexed files under the scoped paths
+- `gcode codewiki --since <git-ref> --out <vault>` — use `git diff --name-only <ref>` to drive incremental regeneration
+- `gcode codewiki --repair-citations --out <vault>` — re-anchor existing `[file:line]` citations without generation or AI calls
+- `gcode codewiki --purge --out <vault> --force` — remove generated CodeWiki Markdown and metadata, then exit; use before a clean rebuild
+
+Aggregate CodeWiki pages may use daemon-side agentic/tool-backed generation when configured. Purge is output-only: it does not clear PostgreSQL code facts, FalkorDB graph data, or Qdrant vectors.
+
 ## When to use which
 
 | Looking for... | Use |

crates/gcode/assets/SKILL.md

@@ -1,6 +1,6 @@
 [package]
 name = "gobby-core"
-version = "0.6.9"
+version = "0.7.0"
 edition = "2024"
 rust-version = "1.88"
 authors = ["Josh Wilhelmi <hello@gobby.ai>"]

crates/gcore/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "gobby-hooks"
-version = "0.6.2"
+version = "0.7.0"
 edition = "2024"
 rust-version = "1.88"
 authors = ["Josh Wilhelmi <hello@gobby.ai>"]
@@ -22,7 +22,7 @@ base64 = "0.22"
 chrono = { version = "0.4", default-features = false, features = ["clock", "std"] }
 clap = { version = "4", features = ["derive"] }
 dirs = "6"
-gobby-core = { path = "../gcore", version = "0.6.1" }
+gobby-core = { path = "../gcore", version = "0.7.0" }
 libc = "0.2"
 log = "0.4"
 serde = { version = "1", features = ["derive"] }

crates/ghook/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "gobby-wiki"
-version = "0.6.7"
+version = "0.7.0"
 edition = "2024"
 rust-version = "1.88"
 authors = ["Josh Wilhelmi <hello@gobby.ai>"]
@@ -26,7 +26,7 @@ name = "gwiki"
 path = "src/main.rs"
 
 [dependencies]
-gobby-core = { path = "../gcore", version = "0.6.1", features = ["postgres", "falkor", "qdrant", "indexing", "search", "graph-analytics", "ai"] }
+gobby-core = { path = "../gcore", version = "0.7.0", features = ["postgres", "falkor", "qdrant", "indexing", "search", "graph-analytics", "ai"] }
 clap = { version = "4", features = ["derive"] }
 anyhow = "1"
 chrono = { version = "0.4", default-features = false, features = ["clock"] }
@@ -70,5 +70,5 @@ libc = "0.2"
 windows-sys = { version = "0.61", features = ["Win32_Foundation", "Win32_Storage_FileSystem"] }
 
 [dev-dependencies]
-gobby-core = { path = "../gcore", version = "0.6.1", features = ["postgres", "falkor", "qdrant", "indexing", "search", "graph-analytics", "ai"] }
+gobby-core = { path = "../gcore", version = "0.7.0", features = ["postgres", "falkor", "qdrant", "indexing", "search", "graph-analytics", "ai"] }
 serial_test = "3"

crates/gwiki/Cargo.toml

@@ -75,7 +75,14 @@ sends a heavier profile for aggregate docs (see the
 
 For direct authenticated endpoints, store the token in user-local
 `~/.gobby/gcore.yaml` as `api_key`. Daemon-backed `config_store` values may use
-`$secret:` references where the daemon runtime config supports them.
+`$secret:` references where the daemon runtime config supports them. CLI-only
+env config sources reject `$secret:` placeholders instead of passing unresolved
+secret names to providers.
+
+Daemon-side agentic generation uses the same profile routing plus a tool policy
+from the caller. `gcode codewiki` uses that path for aggregate handbook pages so
+the daemon agent can investigate through read-only gcode tools before writing;
+direct routes use the local provider-neutral tool loop when configured.
 
 ## Mixed Routing Example
 
@@ -143,4 +150,4 @@ frames, transcription, and text calls share one limiter. Use `ai.keep_alive`
 only with providers that understand it; do not send local-provider-only fields
 to cloud OpenAI-compatible endpoints unless that endpoint documents support.
 
-_Last verified: 2026-06-23_
+_Last verified: 2026-07-01_

docs/guides/ai-configuration.md

@@ -40,6 +40,12 @@ Repair stale `[file:line]` citations in an existing wiki without any AI calls:
 gcode codewiki --repair-citations
 ```
 
+Remove generated CodeWiki Markdown and metadata before a clean rebuild:
+
+```bash
+gcode codewiki --purge --out /path/to/vault --force
+```
+
 ## Output Tree
 
 The generated tree is hierarchical. It has three layers: a **structural layer**
@@ -89,7 +95,9 @@ emitted (so a run with no SystemModel or no deprecations leaves no dead links).
 Metadata:
 
 - `_meta/codewiki.json` records the docs written in the last run for
-  incremental regeneration.
+  incremental regeneration, including each page's render version. Render
+  versions are tracked by page category so a format change to architecture
+  pages does not invalidate unrelated file pages.
 - `_meta/ownership.json` records ownership data for repeat runs.
 - `_meta/truth_digest.json` is the truth digest — a compact, machine-readable
   set of grounded structural facts (see Truth Digest below). Written on full
@@ -125,11 +133,17 @@ The generator then builds docs bottom-up:
    catalog, deprecations) are emitted from the SystemModel, the pinned CLI
    contract, and the deprecation scan — no LLM calls.
 
-When `ai.text_generate` is configured, gcode calls the shared
-`text_generate` route for generated prose. Generated text is citation-checked
-before write, and — when a verifier profile is configured — passes a grounded
-verification step (see Grounded Verification below). Empty or unavailable
-generation falls back to structural AST-only prose.
+When `ai.text_generate` is configured, gcode calls the shared generation route
+for generated prose. Leaf pages use bounded one-shot generation. Aggregate
+handbook pages can use the daemon-side agentic/tool-backed route, where the
+model investigates through read-only gcode tools before answering and the page
+frontmatter records lane, tool-call count, and turn count. Generated text is
+citation-checked before write, normalized through the strict CodeWiki Markdown
+normalizer, and — when a verifier profile is configured — passes a grounded
+verification step (see Grounded Verification below). Empty or unavailable leaf
+generation falls back to structural AST-only prose; failed agentic aggregate
+generation is treated as a hard failure so unsupported handbook pages do not
+silently degrade into skeleton text.
 
 `--ai-depth` controls how deep AI prose generation reaches; gated tiers use
 structural fallbacks:
@@ -141,11 +155,11 @@ structural fallbacks:
   indexed symbol and can take hours-to-days on large repos with local models;
   reserve it for small repos or scoped runs.
 
-On the daemon route, aggregate docs (architecture, module, and repo prose)
-request a heavier daemon profile because they synthesize many child summaries
-into one long grounded answer; file and symbol docs stay on the daemon default
-profile. The aggregate writer chain defaults to an opus-first profile; override
-it with `--ai-aggregate-profile <PROFILE>`.
+On the daemon route, aggregate docs (architecture, module, repo, curated
+concept, and narrative prose) request a heavier daemon profile because they
+synthesize many child summaries into one long grounded answer; file and symbol
+docs stay on the daemon default profile. The aggregate writer chain defaults to
+an opus-first profile; override it with `--ai-aggregate-profile <PROFILE>`.
 
 ### Prose Depth and Audience Register
 
@@ -253,7 +267,27 @@ If the verifier is routed off, unavailable, or returns a malformed verdict, the
 pass is skipped cleanly and the page is emitted undegraded.
 
 `--ai-verify-profile <PROFILE>` (default `feature_mid`) selects the daemon
-profile used for this pass.
+profile used for this pass. `--ai-verify-scope aggregates|all` controls which
+pages run verification: `aggregates` is the default and verifies curated and
+handbook pages only; `all` also verifies per-file leaves and is substantially
+slower on large repos.
+
+## Purge Flow
+
+`gcode codewiki --purge --out <DIR> --force` is a destructive output cleanup
+mode. It removes generated CodeWiki Markdown under `code/` and CodeWiki metadata
+under `_meta/`, then exits. It does not touch raw wiki sources, authored
+knowledge pages, PostgreSQL index rows, FalkorDB graph data, or Qdrant vectors.
+
+`--purge` conflicts with generation, repair, scope, AI, and graph flags; only
+`--out`, `--format`, and `--force` are relevant. Use it when you want to discard
+generated files before a full rebuild:
+
+```bash
+gcode codewiki --purge --out /path/to/vault --force
+gcode codewiki --out /path/to/vault
+gwiki --project /path/to/project index
+```
 
 ## Truth Digest
 
@@ -333,4 +367,4 @@ Because gcode writes vault-ready Markdown, gwiki only needs to classify and
 index the generated subtree. Re-running codewiki rewrites only changed docs, and
 the next gwiki index run picks up only those changed Markdown files.
 
-_Last verified: 2026-06-23_
+_Last verified: 2026-07-01_