@lawaty/lacode 1.0.3 → 1.2.0

package/CHANGELOG.md

@@ -7,6 +7,105 @@ All notable changes to LaCode are recorded here. The format follows
 Versions come from git release commits; the npm version is bumped in the same
 commit as the changelog entry it describes.
 
+## [1.2.0] — voice, a per-session timeline, and a pool without a cap
+
+### Added
+
+- **Voice control for the TUI.** `lacode-voice` runs a local daemon that captures
+  audio, transcribes it through Groq, and forwards prompts to the TUI over a
+  loopback-only control socket. Wake-word detection is optional: the Vosk and
+  Porcupine engines, the recorder and the audio capture path each degrade
+  gracefully when their native dependency, key or device is missing. The daemon,
+  transport and audio code live in `src/voice/`; the TUI is published as the
+  `./tui` export.
+- **A per-session timeline.** `src/plugins/timeline.ts` summarizes a session's
+  transcript into an append-only JSONL file under
+  `~/.local/share/opencode/lacode/timeline/`, and `/reprocess-timeline` rebuilds
+  it from the full transcript through the `timeline_reprocess` tool. Internal
+  summarizer sessions stay out of notifications.
+- **`npm run score`** probes every cost-zero model the configured providers
+  serve, scores context, output, speed and reliability, and suggests a pool
+  order. Results are written to `~/.config/lacode/model-scores.json`.
+
+### Changed
+
+- **BREAKING: Slot routing is one pass, and the four-slot cap is gone.** A
+  slot's `weight` is now its cap, not half of a two-pass budget: the pool fills
+  each slot to its weight in declaration order in one pass, then overflows to the
+  least-loaded slot. Every line is a slot, and a model may repeat — each line
+  adds seats at its position. A preset tuned for the old two-pass behaviour sees
+  roughly half the concurrency it did before; to restore it, double the weights
+  or list a model again.
+- **The pool hot-reloads a bucket-only edit.** Changing only a slot's `b:` bucket
+  now counts as a change and re-validates the table.
+- **A stated `Retry-After` is honoured up to six hours**, instead of being
+  clamped to the fifteen-minute backoff cap.
+
+### Fixed
+
+- **A voice daemon shutdown waits for the transcription in flight**, and manual
+  recording is bounded at two minutes instead of growing without limit.
+- **The voice control socket stays on loopback** unless a non-local bind is
+  explicitly opted into, and the wake engine treats an empty environment value
+  as unset.
+- **The timeline strips terminal control sequences** before rendering model
+  output, and shares its session state across a double-loaded plugin.
+
+## [1.1.0] — a map contract, feedback, and host credential sync
+
+### Added
+
+- **A one-time feedback hint, the `lacode feedback` command, and GitHub issue
+  templates.** The hint is local only: no telemetry, no analytics, and it sends
+  nothing anywhere. It shows at most once, only when a TUI sink exists, and a
+  toast that fails does not consume it. Opt out with `feedbackHint: false` or
+  `LACODE_NO_FEEDBACK_HINT=1`; the flag is
+  `~/.local/share/opencode/lacode/feedback.json`.
+- **The post-publish host sync now syncs provider credentials.** After
+  `npm publish`, `scripts/sync-hosts.mjs` copies the allowlisted providers from
+  the local opencode auth store (`OC_AUTH_PROVIDERS`, default
+  `opencode-go opencode`) to each host, keeps host-only providers, and writes
+  the remote `auth.json` atomically with a timestamped backup.
+
+### Changed
+
+- **The context map has one entry schema, a per-file budget and a per-file
+  lifecycle.** Every navigation entry is a provenance marker (`[binding]`,
+  `[provisional]` or `[disputed]`), a rule of at most three non-blank lines, and
+  a `— See <path>[ §Section] (verified YYYY-MM-DD)` pointer. No Context /
+  Decision / Consequences / Alternatives blocks. Every file has a hard cap in
+  non-blank lines (architecture 150, contexts 200, conventions 150, workflows
+  150, decisions 120, active 60). The 30-day provisional window generalises the
+  active backlog's 7-day finished-task window; binding and disputed entries are
+  exempt, and an entry whose source is gone is never auto-dropped.
+- **The prompt now lives in `src/lib/cartographer-prompt.ts`,** with the tracked
+  copy at `.opencode/prompts/context.txt`; a test pins equality so a reviewer
+  reads the text the model actually runs.
+
+### Fixed
+
+- **`LACODE_SYNC_DRY_RUN=1` no longer contacts a host.** It used to print the
+  plan and then SSH to every host to exercise the credential helper when a local
+  auth store existed, which sent provider credentials during a “dry run”. It now
+  reads only the local plan and prints the auth summary; no SSH and no `npm view`.
+- **The post-publish host sync accepts a real `opencode.jsonc`.** Comments and
+  trailing commas no longer make a host fail as “config not normalized
+  (unparseable)”.
+- **A constraints file is no longer a deliberation ledger.** `decisions.md` is
+  capped, decision-record blocks are a schema violation, and the prompt tells
+  the cartographer to state the constraint and point at the page that owns it.
+- **Restatement and drift are now reported instead of silently trusted.** After
+  every run the map is audited for dangling paths/sections, a numeric claim
+  absent from its cited source, a possible negation conflict, a source modified
+  after `verified`, tables/fenced blocks/verbatim spans, duplicate rules,
+  archive-of-record language and charter degeneration. The audit is
+  **report-only**: it never rewrites a claim and never deletes content with no
+  source, because the map may be the only copy and the page may be the stale
+  side. A failure is a loud `context map validation failed` error.
+- **The lifecycle no longer belongs to one file.** `active.md` still uses its
+  seven-day finished-task window; every other file now expires provisional
+  entries after 30 days, and only when the cited source still resolves.
+
 ## [1.0.3] — one notification even when the plugin is loaded twice
 
 ### Fixed
@@ -820,7 +919,9 @@ experimental; `latest` remains 0.4.2.
 
 - First published release: pooled subagents over a shared set of models.
 
-[Unreleased]: https://github.com/lawaty/lacode/compare/v1.0.0...HEAD
+[Unreleased]: https://github.com/lawaty/lacode/compare/v1.2.0...HEAD
+[1.2.0]: https://github.com/lawaty/lacode/compare/v1.1.0...v1.2.0
+[1.1.0]: https://github.com/lawaty/lacode/compare/v1.0.0...v1.1.0
 [1.0.0]: https://github.com/lawaty/lacode/compare/v0.11.0...v1.0.0
 [0.11.0]: https://github.com/lawaty/lacode/compare/v0.10.0...v0.11.0
 [0.10.0]: https://github.com/lawaty/lacode/compare/v0.9.0...v0.10.0

package/README.md

@@ -104,6 +104,23 @@ In a repository with no `.opencode/context/` yet, the first session builds it in
 the background; `/context-init` forces it, and [Context map](#context-map)
 explains what it is for.
 
+## Feedback
+
+The feedback path has no telemetry and no analytics — it sends nothing anywhere
+— but I would still like to hear from you.
+
+- **Bug or feature request?** [Open an issue](https://github.com/lawaty/lacode/issues).
+- **Question, or want to say how you use it?** [Start a discussion](https://github.com/lawaty/lacode/discussions).
+- **Found it useful?** A ⭐ on [the repo](https://github.com/lawaty/lacode) helps
+  other people find it.
+
+`lacode feedback` prints both links. The one-time hint that points at the issue
+tracker is opt-out: set `feedbackHint: false` in the plugin options, or
+`LACODE_NO_FEEDBACK_HINT=1` in the environment. It is shown at most once, only
+when there is a TUI to show it in; piped, CI and test runs leave the flag
+untouched. A toast that fails does not consume the note, so a later interactive
+run can still show it.
+
 ## Configure
 
 Plugin options are a tuple — the package name, then an options object. Every
@@ -128,11 +145,12 @@ option is optional:
 | Option | Default | Meaning |
 | --- | --- | --- |
 | `enabled` | `true` | The single off switch. `false` injects nothing at all and does no writes. |
-| `models` | — | An inline slot list. Highest-priority source. |
+| `models` | — | An inline slot list; `weight` is that line's budget and cap, a model may repeat, and there is no slot cap. Highest-priority source. |
 | `modelsFile` | — | An explicit slot file; short-circuits the [resolution chain](docs/CONFIGURATION.md#pool-preset-resolution). |
 | `reapMs` | `30000` | Hang-reaper interval. |
 | `probeMs` | `900000` | How often benched models are checked by actually using them. `0` turns the periodic check off and leaves the `pool_recheck` tool as the only way to ask. |
 | `selfUpdate` | `true` | Check for a newer release at most once per ~6h and install it in the background; it applies on the next opencode start. `false` — or `LACODE_NO_SELFUPDATE=1` — turns the check off. |
+| `feedbackHint` | `true` | Show the one-time feedback note on the first interactive run. `false` — or `LACODE_NO_FEEDBACK_HINT=1` — turns it off. The flag lives in `~/.local/share/opencode/lacode/feedback.json`; nothing is sent anywhere. |
 | `sound` | `true` | Play a short tone with each notification, one per outcome, so you can tell a finished session from a question or a permission prompt without reading the popup. It is opt-out: `false` — or `LACODE_SOUND=0` (`false`, `off`, `no` work too, and beat the option) — turns the sound off, and anything else, a typo included, leaves it on. |
 | `omniroute` | — | Optional [OmniRoute](docs/OMNIROUTE.md) gateway integration. Absent or `{ enabled: false }` changes nothing. |
 | `watchMs` | `1000` | How often the pool files are checked for changes. `0` disables live reload. |
@@ -195,11 +213,17 @@ When it reports `stale`, rebuild with `npm run build`. No network is touched;
 ## What you get
 
 Three pooled agents: `explore`, `implement` and `context`. You spawn the base
-name and the pool picks a cheap model slot for it. Behind each base sits up to
-four hidden variants — one per slot — and a spawn of the bare name is the only
-thing that works: a hand-named variant such as `explore-1` is coerced to its
-base and routed by the pool, so naming one cannot pin a provider or skip
-ordered slot filling.
+name and the pool picks a cheap model slot for it. Behind each base sits one
+hidden variant per slot — however many the pool declares — and a spawn of the
+bare name is the only thing that works: a hand-named variant such as `explore-1`
+is coerced to its base and routed by the pool, so naming one cannot pin a
+provider or skip ordered slot filling.
+
+A slot's `weight` is both its concurrency budget and its cap: the pool fills
+slots in declaration order in one pass, then overflows to the least-loaded slot.
+There is no slot cap, and a model may appear on more than one line — each line
+is a separate slot with its own weight — so to give a model more capacity, list
+it again.
 
 A model that rate-limits or runs out of quota is benched until its window ends,
 so one sick provider does not take the pool with it.
@@ -217,9 +241,9 @@ serve at all is benched for the quota cap and named in the report.
 A change is reported as a toast and a log line; a pass that finds nothing changed says
 nothing at all.
 
-Nine commands: `/context-init`, `/context-update`, `/context-review`,
-`/lacode-doctor`, `/agent-instructions`, `/pool`, `/pool-recheck`, `/image-in-cb`
-and `/image`. Everything is added **only when your config does not already
+Ten commands: `/context-init`, `/context-update`, `/context-review`,
+`/lacode-doctor`, `/agent-instructions`, `/pool`, `/pool-recheck`,
+`/reprocess-timeline`, `/image-in-cb` and `/image`. Everything is added **only when your config does not already
 define it** — an agent or command you wrote yourself is left byte-identical,
 never patched or partially merged. `{"enabled": false}` turns the whole thing off
 without leaving a trace. The exact injection list, and the five keys that are
@@ -373,6 +397,12 @@ to emit a number they cannot derive — an unpriced model reports `unknown`, nev
 `0`. How the two modes differ is in [docs/BENCHMARKS.md](docs/BENCHMARKS.md),
 which explains the harness and deliberately contains no results.
 
+`npm run score` is the other measurement: it probes each configured provider's
+free models with one tiny request and ranks them by measured latency, context and
+output limit. It prints a suggested `pool.txt` but never edits it, and writes the
+raw rows to `~/.config/lacode/model-scores.json`. Run it every few days — free
+models and quota windows move.
+
 ## Remote hosts
 
 Notifications raised on a remote host need a way back to your desktop. The
@@ -393,12 +423,13 @@ and **Deny**, and macOS gets a plain notice without buttons.
 `lacode forward HOST -- COMMAND` runs a command on another host over ssh and
 exits with its status.
 
-The full CLI is `notify`, `forward`, `doctor`, `version`, `update`, `receive`
-and `tunnel setup`. `version` reports which copy of LaCode is running — and
-whether opencode is loading that same copy — and `update [--check]` refreshes an
-installed one. `receive` is the desktop end of the tunnel (the forced command a
-remote key is restricted to) and `tunnel setup` provisions the key and the
-forward.
+The full CLI is `notify`, `forward`, `doctor`, `feedback`, `version`, `update`,
+`receive` and `tunnel setup`. `version` reports which copy of LaCode is running —
+and whether opencode is loading that same copy — and `update [--check]` refreshes
+an installed one. `feedback` prints the Issues and Discussions links; it opens
+nothing and sends nothing. `receive` is the desktop end of the tunnel (the forced
+command a remote key is restricted to) and `tunnel setup` provisions the key and
+the forward.
 
 Configuration, the payload format, the receiver contract and holding the forward
 open are in [docs/TUNNEL.md](docs/TUNNEL.md).
@@ -426,6 +457,7 @@ want it.
 npm run build          # esbuild -> dist/ (what the tarball ships)
 npm test               # node --test tests/ — offline, no network, no running server
 npm run benchmark      # the cost harness; see docs/BENCHMARKS.md
+npm run score          # rank free models by context, output and measured latency
 ```
 
 Node ≥ 22.18.