kaizen-agent 1.7.1 → 1.7.3

package/skills/kaizen-help/SKILL.md

@@ -1,129 +1,129 @@
----
-name: kaizen-help
-description: >
-  Quick-reference card for the kaizen workflow: commands, modes, approvals,
-  fix loop, config knobs, state layout, backlog/status rules. One-shot
-  display, not a persistent mode. Trigger: /kaizen-help, "kaizen help",
-  "what kaizen commands", "kaizen cheat sheet".
----
-
-# Kaizen Help
-
-**Print the content of this card into your reply.** The user cannot see this file —
-it loads into your context only, so summarizing it or saying "see above" shows them
-nothing. Output the tables themselves.
-
-One-shot: do NOT change kaizen's active mode, write to `state.json`, or persist
-anything.
-
-## Commands
-
-| Invocation | Action |
-|---|---|
-| `/kaizen <request>` | Start a new run in the configured mode (default: `approve`) |
-| `/kaizen plan <request>` | Start a run and stop after the plan (`plan-only`) |
-| `/kaizen auto <request>` | Start a run in `auto` mode: every stage back to back, no approval stops |
-| `/kaizen lite <request>` | Run the stages in this session — cheap, reviewer has seen the work |
-| `/kaizen full <request>` | Dispatch each stage as its own cold agent |
-| `/kaizen run` | Resume the current run and execute the approved plan |
-| `/kaizen review [target]` | Review-only: audit existing code, no plan, no implementation |
-| `/kaizen status` | List every run grouped waiting-on-you / in flight / done / abandoned, plus open backlog count |
-| `/kaizen backlog` | Print open backlog items across all runs, grouped by source run, rescued items last under `Orphaned` |
-| `/kaizen approve` | Approve whatever the current run is waiting on |
-| `/kaizen reject <reason>` | Reject it; the reason decides revise (back to planner) vs kill (abandon) |
-| `/kaizen abort` | Mark the current run abandoned |
-| `/kaizen config` | Show this project's settings and change them, one picker per setting |
-| `/kaizen init` | Set up this project folder: `.kaizen/` with `spec.md`, config, and — in a git repo — the gitignore entry |
-| `/kaizen install` | Install the workflow itself globally or per-project |
-
-`/kaizen` with no argument: `status` if a run is in progress, else ask what to build.
-
-## Modes
-
-Set by `mode` in config, overridable per invocation.
-
-| Mode | What it does |
-|---|---|
-| `plan-only` | Produce the plan, write it, stop. No code touched. |
-| `approve` (default) | Stop at every enabled approval, wait for an explicit decision. |
-| `auto` | Run every stage back to back, no approval, report at the end. Fix loop still respects `max_iterations`. |
-| `review-only` | Skip plan and implementation, run the reviewer against existing code or a diff. |
-
-Mode never overrides a refusal: destructive/irreversible actions still confirm even in `auto`.
-
-## Approvals
-
-- `approvals.plan` — after the plan, before any code is written.
-- `approvals.review` — after the review passes, before the work is called done.
-- `approvals.each_file` — confirm each file edit individually. Off by default; slow.
-
-A rejection is written to the run directory, and is one of two things — the reason
-says which. **Revise** ("do it differently") goes back to the planner, which produces
-a new plan and returns to the same approval. **Kill** ("not doing this") means the
-request itself was wrong: set `stage: "abandoned"` and stop, nothing to revise toward.
-
-## Fix loop
-
-- `auto_fix.enabled: false` — every finding stops the run, becomes a user decision.
-- `auto_fix.enabled: true` — builder fixes findings at or above `auto_fix.min_severity`,
-  reviewer re-checks, up to `auto_fix.max_iterations` rounds. Findings below threshold,
-  and anything still open when iterations run out, are reported to the user.
-
-Each iteration is written to `runs/<id>/05-iterations/`.
-
-## Config knobs (`config.default.yml` defaults)
-
-| Key | Default | Meaning |
-|---|---|---|
-| `preset` | `medium` | Quick preset (`low`, `medium`, `ultra`, `custom`) |
-| `mode` | `approve` | Default run mode |
-| `runner` | `full` | `full`: each stage a cold subagent. `lite`: this session runs every stage |
-| `approvals.plan` | `true` | Stop after the plan |
-| `approvals.review` | `false` | Stop after the review |
-| `approvals.each_file` | `false` | Confirm every file edit |
-| `build.executor` | `subagent` | Who builds: `subagent`, `inline` (main thread), or `ask` at the approval |
-| `auto_fix.enabled` | `true` | Auto-fix findings vs. escalate everything |
-| `auto_fix.min_severity` | `high` | Threshold for auto-fix (`critical\|high\|medium\|low`) |
-| `auto_fix.max_iterations` | `2` | Fix/recheck rounds before escalating remaining findings |
-| `track.default` | `auto` (inferred per run) | Pin a track for a repo that is always one kind of work |
-| `review.write_memory` | `true` | Let the reviewer append lessons to `.kaizen/memory.md` |
-| `backlog.enabled` | `true` | Capture what a run deliberately did not do |
-| `backlog.file` | `06-backlog.md` | Per-run backlog file |
-| `backlog.orphan_file` | `backlog.md` | Where a pruned run's open items are rescued to |
-| `state.dir` | `.kaizen` | State directory |
-| `state.keep_runs` | `20` | Completed runs kept before pruning oldest (rescues open backlog first) |
-| `git.auto_commit` | `false` | Never commit/push/PR unless the user explicitly asks |
-| `git.branch_before_implement` | `true` | Branch before implement stage when on the default branch |
-| `subagents.model` | `inherit` | Model the stage subagents run on (opus/sonnet/haiku, Claude Code only) |
-| `subagents.effort` | `inherit` | Reasoning effort for the stage subagents (low/medium/high, Claude Code only) |
-| `agent.default` | `auto` | Coding agent to launch from dashboard backlog |
-| `agent.<cmd>.model`, `.effort` | — | Model/effort flags for the launched session, per tool (`claude`, `codex`, `agy`, `opencode`, `gemini`); model names are the tool's own |
-
-`.kaizen/config.yml` overrides only the keys it sets; everything else falls back to
-this skill's `config.default.yml`.
-
-## State layout
-
-```
-.kaizen/{spec.md, config.yml, memory.md, backlog.md (orphanage)}
-.kaizen/runs/<id>/{00-request,01-plan,02-approval,03-impl,04-review}.md,
-  05-iterations/, 06-backlog.md, state.json (id, mode, stage, awaiting, iteration, updated)
-```
-
-## Backlog and status
-
-Backlog items: `- <status>: <item text>`, `<status>` one of `open`, `done`, `rejected`
-(`rejected` carries a reason after `|`). `/kaizen backlog`/`status` are **read-only** —
-never edit `06-backlog.md` or `state.json`, even to fix a spotted inconsistency; report
-it and ask.
-
-`/kaizen status` groups every run by `state.json` at read time: **Waiting on you**
-(awaiting non-null, not abandoned), **In flight** (awaiting null, not done/abandoned),
-**Done**, **Abandoned**, plus `Backlog: <n> open` (all runs + orphanage; abandoned
-runs' items excluded).
-
-## More
-
-Full walkthrough and rationale: `~/.claude/skills/kaizen/README.md`. Full stage
-contract: `~/.claude/skills/kaizen/spec.md`. Codex/Antigravity porting:
-`~/.claude/skills/kaizen/adapters.md`.
+---
+name: kaizen-help
+description: >
+  Quick-reference card for the kaizen workflow: commands, modes, approvals,
+  fix loop, config knobs, state layout, backlog/status rules. One-shot
+  display, not a persistent mode. Trigger: /kaizen-help, "kaizen help",
+  "what kaizen commands", "kaizen cheat sheet".
+---
+
+# Kaizen Help
+
+**Print the content of this card into your reply.** The user cannot see this file —
+it loads into your context only, so summarizing it or saying "see above" shows them
+nothing. Output the tables themselves.
+
+One-shot: do NOT change kaizen's active mode, write to `state.json`, or persist
+anything.
+
+## Commands
+
+| Invocation | Action |
+|---|---|
+| `/kaizen <request>` | Start a new run in the configured mode (default: `approve`) |
+| `/kaizen plan <request>` | Start a run and stop after the plan (`plan-only`) |
+| `/kaizen auto <request>` | Start a run in `auto` mode: every stage back to back, no approval stops |
+| `/kaizen lite <request>` | Run the stages in this session — cheap, reviewer has seen the work |
+| `/kaizen full <request>` | Dispatch each stage as its own cold agent |
+| `/kaizen run` | Resume the current run and execute the approved plan |
+| `/kaizen review [target]` | Review-only: audit existing code, no plan, no implementation |
+| `/kaizen status` | List every run grouped waiting-on-you / in flight / done / abandoned, plus open backlog count |
+| `/kaizen backlog` | Print open backlog items across all runs, grouped by source run, rescued items last under `Orphaned` |
+| `/kaizen approve` | Approve whatever the current run is waiting on |
+| `/kaizen reject <reason>` | Reject it; the reason decides revise (back to planner) vs kill (abandon) |
+| `/kaizen abort` | Mark the current run abandoned |
+| `/kaizen config` | Show this project's settings and change them, one picker per setting |
+| `/kaizen init` | Set up this project folder: `.kaizen/` with `spec.md`, config, and — in a git repo — the gitignore entry |
+| `/kaizen install` | Install the workflow itself globally or per-project |
+
+`/kaizen` with no argument: `status` if a run is in progress, else ask what to build.
+
+## Modes
+
+Set by `mode` in config, overridable per invocation.
+
+| Mode | What it does |
+|---|---|
+| `plan-only` | Produce the plan, write it, stop. No code touched. |
+| `approve` (default) | Stop at every enabled approval, wait for an explicit decision. |
+| `auto` | Run every stage back to back, no approval, report at the end. Fix loop still respects `max_iterations`. |
+| `review-only` | Skip plan and implementation, run the reviewer against existing code or a diff. |
+
+Mode never overrides a refusal: destructive/irreversible actions still confirm even in `auto`.
+
+## Approvals
+
+- `approvals.plan` — after the plan, before any code is written.
+- `approvals.review` — after the review passes, before the work is called done.
+- `approvals.each_file` — confirm each file edit individually. Off by default; slow.
+
+A rejection is written to the run directory, and is one of two things — the reason
+says which. **Revise** ("do it differently") goes back to the planner, which produces
+a new plan and returns to the same approval. **Kill** ("not doing this") means the
+request itself was wrong: set `stage: "abandoned"` and stop, nothing to revise toward.
+
+## Fix loop
+
+- `auto_fix.enabled: false` — every finding stops the run, becomes a user decision.
+- `auto_fix.enabled: true` — builder fixes findings at or above `auto_fix.min_severity`,
+  reviewer re-checks, up to `auto_fix.max_iterations` rounds. Findings below threshold,
+  and anything still open when iterations run out, are reported to the user.
+
+Each iteration is written to `runs/<id>/05-iterations/`.
+
+## Config knobs (`config.default.yml` defaults)
+
+| Key | Default | Meaning |
+|---|---|---|
+| `preset` | `medium` | Quick preset (`low`, `medium`, `ultra`, `custom`) |
+| `mode` | `approve` | Default run mode |
+| `runner` | `full` | `full`: each stage a cold subagent. `lite`: this session runs every stage |
+| `approvals.plan` | `true` | Stop after the plan |
+| `approvals.review` | `false` | Stop after the review |
+| `approvals.each_file` | `false` | Confirm every file edit |
+| `build.executor` | `subagent` | Who builds: `subagent`, `inline` (main thread), or `ask` at the approval |
+| `auto_fix.enabled` | `true` | Auto-fix findings vs. escalate everything |
+| `auto_fix.min_severity` | `high` | Threshold for auto-fix (`critical\|high\|medium\|low`) |
+| `auto_fix.max_iterations` | `2` | Fix/recheck rounds before escalating remaining findings |
+| `track.default` | `auto` (inferred per run) | Pin a track for a repo that is always one kind of work |
+| `review.write_memory` | `true` | Let the reviewer append lessons to `.kaizen/memory.md` |
+| `backlog.enabled` | `true` | Capture what a run deliberately did not do |
+| `backlog.file` | `06-backlog.md` | Per-run backlog file |
+| `backlog.orphan_file` | `backlog.md` | Where a pruned run's open items are rescued to |
+| `state.dir` | `.kaizen` | State directory |
+| `state.keep_runs` | `20` | Completed runs kept before pruning oldest (rescues open backlog first) |
+| `git.auto_commit` | `false` | Never commit/push/PR unless the user explicitly asks |
+| `git.branch_before_implement` | `true` | Branch before implement stage when on the default branch |
+| `subagents.model` | `inherit` | Model the stage subagents run on (opus/sonnet/haiku, Claude Code only) |
+| `subagents.effort` | `inherit` | Reasoning effort for the stage subagents (low/medium/high, Claude Code only) |
+| `agent.default` | `auto` | Coding agent to launch from dashboard backlog |
+| `agent.<cmd>.model`, `.effort` | — | Model/effort flags for the launched session, per tool (`claude`, `codex`, `agy`, `opencode`, `gemini`); model names are the tool's own |
+
+`.kaizen/config.yml` overrides only the keys it sets; everything else falls back to
+this skill's `config.default.yml`.
+
+## State layout
+
+```
+.kaizen/{spec.md, config.yml, memory.md, backlog.md (orphanage)}
+.kaizen/runs/<id>/{00-request,01-plan,02-approval,03-impl,04-review}.md,
+  05-iterations/, 06-backlog.md, state.json (id, mode, stage, awaiting, iteration, updated)
+```
+
+## Backlog and status
+
+Backlog items: `- <status>: <item text>`, `<status>` one of `open`, `done`, `rejected`
+(`rejected` carries a reason after `|`). `/kaizen backlog`/`status` are **read-only** —
+never edit `06-backlog.md` or `state.json`, even to fix a spotted inconsistency; report
+it and ask.
+
+`/kaizen status` groups every run by `state.json` at read time: **Waiting on you**
+(awaiting non-null, not abandoned), **In flight** (awaiting null, not done/abandoned),
+**Done**, **Abandoned**, plus `Backlog: <n> open` (all runs + orphanage; abandoned
+runs' items excluded).
+
+## More
+
+Full walkthrough and rationale: `~/.claude/skills/kaizen/README.md`. Full stage
+contract: `~/.claude/skills/kaizen/spec.md`. Codex/Antigravity porting:
+`~/.claude/skills/kaizen/adapters.md`.

package/web/board.html

@@ -27,6 +27,8 @@ pre.mark span:nth-child(1){color:var(--ink-hi)}pre.mark span:nth-child(2){color:
 .ib.pri{color:var(--accent)}
 .ib.sm{width:26px;height:26px}.ib.sm svg{width:15px;height:15px}
 /* dictation mic: accent while listening (pulsing) or still transcribing; digits while the model downloads */
+/* the field being dictated into: accent border and a glow that swells with the voice level (--lvl, 0..1) */
+.dictating{border-color:var(--accent)!important;box-shadow:0 0 0 calc(2px + 6px*var(--lvl,0)) color-mix(in srgb,var(--accent) 35%,transparent);transition:box-shadow .08s}
 .ib.mic.on{color:var(--accent);animation:pl 1.6s infinite}.ib.mic.busy{color:var(--accent)}.ib.mic i{font-style:normal;font-size:10px;font-variant-numeric:tabular-nums}.row .mic{margin-right:auto;align-self:center}
 .btn{border:1px solid var(--line);border-radius:var(--r);padding:5px 12px;color:var(--ink);background:var(--bg-2);min-height:30px;font-size:13px}
 .btn:hover{border-color:var(--dim)}
@@ -156,7 +158,7 @@ aside .foot:empty{display:none}
 .help.on{display:flex}.help div{background:var(--bg-2);border:1px solid var(--line);border-radius:var(--r);padding:20px 24px;line-height:2}
 .help kbd{display:inline-block;width:7ch;color:var(--accent)}
 @media(max-width:900px){main{grid-template-columns:1fr;padding-inline:16px;gap:8px}.hdr,.strip,.ftr{padding-inline:16px}section h2{position:sticky;top:0;background:var(--bg);z-index:1}.more{display:grid}.card:hover .age{display:inline}}
-@media(prefers-reduced-motion:reduce){aside,.toast,.mascot,.d-run .pulse,.ib.mic.on{transition:none;animation:none}}
+@media(prefers-reduced-motion:reduce){aside,.toast,.mascot,.d-run .pulse,.ib.mic.on,.dictating{transition:none;animation:none}}
 </style>
 </head>
 <body>
@@ -229,7 +231,12 @@ const TFJS="https://cdn.jsdelivr.net/npm/@huggingface/transformers@4.3.0",ASR_MO
 // Inference lives in a module Worker so a slow chunk never freezes the board.
 const WORKER_SRC=`import{pipeline}from"${TFJS}";let asr=null;
 onmessage=async e=>{const m=e.data;try{
- if(m.type==="load"){asr=await pipeline("automatic-speech-recognition","${ASR_MODEL}",{device:m.webgpu?"webgpu":"wasm",dtype:m.webgpu?{encoder_model:"fp16",decoder_model_merged:"q4"}:"q8",progress_callback:p=>{if(p.status==="progress")postMessage({type:"progress",file:p.file,pct:p.progress||0})}});postMessage({type:"ready"})}
+ if(m.type==="load"){const prog=p=>{if(p.status==="progress")postMessage({type:"progress",file:p.file,pct:p.progress||0})};
+  // WebGPU first, with the encoder in fp16 only where the adapter has shader-f16; any GPU failure falls back to the CPU.
+  const gpu=m.webgpu?{device:"webgpu",dtype:{encoder_model:m.f16?"fp16":"fp32",decoder_model_merged:"q4"}}:null,cpu={device:"wasm",dtype:"q8"};
+  try{if(!gpu)throw 0;asr=await pipeline("automatic-speech-recognition","${ASR_MODEL}",{...gpu,progress_callback:prog})}
+  catch(e){if(gpu)postMessage({type:"fallback",why:String(e&&e.message||e)});asr=await pipeline("automatic-speech-recognition","${ASR_MODEL}",{...cpu,progress_callback:prog})}
+  postMessage({type:"ready"})}
  if(m.type==="run"){const r=await asr(m.audio,{chunk_length_s:30,return_timestamps:false});postMessage({type:"text",sid:m.sid,seq:m.seq,text:(r.text||"").trim()})}
 }catch(err){postMessage({type:"error",sid:m.sid,seq:m.seq,error:String(err&&err.message||err)})}}`;
 // The worklet only forwards 128-frame blocks; chunking and silence detection stay on the main thread.
@@ -238,11 +245,13 @@ const HAS_MIC=!!(navigator.mediaDevices&&navigator.mediaDevices.getUserMedia&&wi
 // `active` is the one session listening; `sessions` also holds stopped ones still waiting on a chunk.
 let asrWorker=null,asrReady=null,asrProgress=()=>{},active=null,starting=false;const sessions=new Set();
 // One worker for the page; loading is awaited by whoever pressed first, later presses join.
-function loadAsr(){if(asrReady)return asrReady;asrReady=new Promise((res,rej)=>{
+function loadAsr(){if(asrReady)return asrReady;asrReady=new Promise(async(res,rej)=>{
  asrWorker=new Worker(URL.createObjectURL(new Blob([WORKER_SRC],{type:"text/javascript"})),{type:"module"});
- const files={};asrWorker.onmessage=e=>{const m=e.data;if(m.type==="progress"){files[m.file]=m.pct;const v=Object.values(files);asrProgress(Math.round(v.reduce((a,b)=>a+b,0)/v.length))}else if(m.type==="ready")res();else if(m.type==="error"&&m.seq==null)rej(new Error(m.error));for(const s of sessions)s.onmsg(m)};
+ const files={};asrWorker.onmessage=e=>{const m=e.data;if(m.type==="progress"){files[m.file]=m.pct;const v=Object.values(files);asrProgress(Math.round(v.reduce((a,b)=>a+b,0)/v.length))}else if(m.type==="ready")res();else if(m.type==="fallback")toast("GPU could not run the model, using the CPU: "+m.why);else if(m.type==="error"&&m.seq==null)rej(new Error(m.error));for(const s of sessions)s.onmsg(m)};
  asrWorker.onerror=e=>rej(new Error(e.message||"worker failed"));
- const webgpu=!!navigator.gpu;if(!webgpu)toast("No WebGPU: dictation will run on the CPU and be slower");asrWorker.postMessage({type:"load",webgpu})});
+ // Ask the adapter what it can do: no adapter means the CPU path; no shader-f16 means an fp32 encoder.
+ let adapter=null;try{adapter=navigator.gpu&&await navigator.gpu.requestAdapter()}catch(e){}
+ if(!adapter)toast("No WebGPU: dictation will run on the CPU and be slower");asrWorker.postMessage({type:"load",webgpu:!!adapter,f16:!!(adapter&&adapter.features.has("shader-f16"))})});
  asrReady.catch(()=>{asrReady=null;try{asrWorker.terminate()}catch(e){}asrWorker=null});return asrReady}
 // After a re-render only a listener whose field is gone is stopped; closing the panel stops any.
 function stopDictation(all){if(active&&(all||!active.field.isConnected))active.stop()}
@@ -272,14 +281,14 @@ function mic(fields){
   const flush=()=>{if(!len||voiced<MIN){buf=[];len=0;voiced=0;quiet=0;return}const a=new Float32Array(len);let o=0;for(const c of buf){a.set(c,o);o+=c.length}buf=[];len=0;voiced=0;quiet=0;pending.set(seq,null);asrWorker.postMessage({type:"run",sid,seq,audio:a},[a.buffer]);seq++;b.classList.add("busy")};
   const drain=()=>{while(pending.has(done)&&pending.get(done)!==null){insert(pending.get(done));pending.delete(done);done++}if(!pending.size){b.classList.remove("busy");if(stopping){idle();sessions.delete(sess)}}};
   tap.port.onmessage=e=>{const c=e.data;buf.push(c);len+=c.length;let s=0;for(let i=0;i<c.length;i++)s+=c[i]*c[i];const rms=Math.sqrt(s/c.length);
-   if(rms>0.01){voiced+=c.length;quiet=0}else quiet+=c.length;
+   if(rms>0.01){voiced+=c.length;quiet=0}else quiet+=c.length;f.style.setProperty("--lvl",Math.min(1,rms*8).toFixed(2));
    if(len>=MAX||(voiced>=MIN&&quiet>=SIL))flush()};
   // Chunk seqs are per session, so a result is matched by session id too.
   const sid=Math.random().toString(36).slice(2);
   const sess={btn:b,field:f,onmsg:m=>{if(m.sid!==sid)return;if(m.type==="text"){pending.set(m.seq,m.text);drain()}else if(m.type==="error"&&m.seq!=null){pending.set(m.seq,"");toast("Dictation failed: "+m.error);drain()}},
-   stop:()=>{if(stopping)return;stopping=true;b.classList.remove("on");try{tap.port.onmessage=null;src.disconnect();tap.disconnect();stream.getTracks().forEach(t=>t.stop());ctx.close()}catch(e){}active=null;flush();drain()}};
+   stop:()=>{if(stopping)return;stopping=true;b.classList.remove("on");f.classList.remove("dictating");f.style.removeProperty("--lvl");try{tap.port.onmessage=null;src.disconnect();tap.disconnect();stream.getTracks().forEach(t=>t.stop());ctx.close()}catch(e){}active=null;flush();drain()}};
   sessions.add(sess);active=sess;starting=false;
-  b.classList.remove("busy");b.classList.add("on");f.focus()};
+  b.classList.remove("busy");b.classList.add("on");f.classList.add("dictating");f.focus()};
  return b}
 const m=60e3,h=3600e3,d=86400e3;
 const age=(t,now)=>{const x=now-t;return x<h?Math.round(x/m)+"m":x<d?Math.round(x/h)+"h":Math.round(x/d)+"d"};