@pmelab/gtd 29.2.0 → 29.2.1

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "gtd",
-  "version": "29.2.0",
+  "version": "29.2.1",
   "description": "Drive the gtd loop inside Claude Code: agent turns run as subagents, gtd decides every beat",
   "author": {
     "name": "Philipp Melab"

package/claude/hooks/judge.ts

@@ -83,7 +83,9 @@ export function judgePrompt(j: Judgment): string {
     "Answer every question below from the evidence alone. For each, give your answer and p, your confidence between 0 and 1 that the given answer is correct.",
   ]
   for (const [key, body] of Object.entries(j.state ?? {}))
-    parts.push(`<evidence key="${key}">\n${body}\n</evidence>`)
+    parts.push(
+      `<evidence key="${key}">\n${body.replace(/<\/evidence/gi, "<\\/evidence")}\n</evidence>`,
+    )
   for (const q of j.questions) {
     const levels = levelsOf(q)
     if (typeof levels === "string") throw new Error(levels)

package/dist/flows.mjs

@@ -370,7 +370,7 @@ const SHAPE_CRITERIA = [
 	"land: every stray is incidental (a needed import, a test fixture, a renamed call site).",
 	"refactor-first: the strays are a seam change the task needs; land them first as a behaviour-neutral refactor.",
 	"reject: the naive pass solved the task the wrong way; discard it and re-implement, no human; strays a refactor could absorb never count as reject."
-].join("\n");
+].join(" ");
 const readable = (c) => (c.status === "deleted" || c.after !== void 0) && (c.status === "added" || c.before !== void 0);
 const strayContent = (strays) => strays.map((c) => [
 	`=== ${c.path} (${c.status})`,

package/dist/gtd.bundle.mjs

@@ -22014,7 +22014,7 @@ const SHAPE_CRITERIA = [
 	"land: every stray is incidental (a needed import, a test fixture, a renamed call site).",
 	"refactor-first: the strays are a seam change the task needs; land them first as a behaviour-neutral refactor.",
 	"reject: the naive pass solved the task the wrong way; discard it and re-implement, no human; strays a refactor could absorb never count as reject."
-].join("\n");
+].join(" ");
 const readable = (c) => (c.status === "deleted" || c.after !== void 0) && (c.status === "added" || c.before !== void 0);
 const strayContent = (strays) => strays.map((c) => [
 	`=== ${c.path} (${c.status})`,
@@ -41238,7 +41238,9 @@ const planLanding = (opts = {}, current) => gen(function* () {
 * closes or sits inside its own closing message. Writes nothing — no git, no
 * state transition, no file. Refuses (throws, mapped to the runtime-error
 * exit code) when the workflow declares no `summary` prompt, or the resolved
-* run has no commits to name. The edge never narrates: the CLI prints the
+* run has no commits to name. `--slice` also refuses an unknown slice name
+* (listing the valid slices), a config edited since the process closed, and a
+* replay that diverged. The edge never narrates: the CLI prints the
 * version warning, before the unknown-slice refusal; a replay refusal carries
 * it in its message.
 */

package/dist/testkit.mjs

@@ -10341,7 +10341,9 @@ const planLanding = (opts = {}, current) => gen(function* () {
 * closes or sits inside its own closing message. Writes nothing — no git, no
 * state transition, no file. Refuses (throws, mapped to the runtime-error
 * exit code) when the workflow declares no `summary` prompt, or the resolved
-* run has no commits to name. The edge never narrates: the CLI prints the
+* run has no commits to name. `--slice` also refuses an unknown slice name
+* (listing the valid slices), a config edited since the process closed, and a
+* replay that diverged. The edge never narrates: the CLI prints the
 * version warning, before the unknown-slice refusal; a replay refusal carries
 * it in its message.
 */

package/dist/workflows.mjs

@@ -349,7 +349,7 @@ const SHAPE_CRITERIA = [
 	"land: every stray is incidental (a needed import, a test fixture, a renamed call site).",
 	"refactor-first: the strays are a seam change the task needs; land them first as a behaviour-neutral refactor.",
 	"reject: the naive pass solved the task the wrong way; discard it and re-implement, no human; strays a refactor could absorb never count as reject."
-].join("\n");
+].join(" ");
 const readable = (c) => (c.status === "deleted" || c.after !== void 0) && (c.status === "added" || c.before !== void 0);
 const strayContent = (strays) => strays.map((c) => [
 	`=== ${c.path} (${c.status})`,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pmelab/gtd",
-  "version": "29.2.0",
+  "version": "29.2.1",
   "private": false,
   "description": "Git-aware CLI that emits the next prompt for an autonomous coding agent based on the current repository state",
   "bin": {

package/src/cli/Program.ts

@@ -326,7 +326,9 @@ const planLanding = (
  * closes or sits inside its own closing message. Writes nothing — no git, no
  * state transition, no file. Refuses (throws, mapped to the runtime-error
  * exit code) when the workflow declares no `summary` prompt, or the resolved
- * run has no commits to name. The edge never narrates: the CLI prints the
+ * run has no commits to name. `--slice` also refuses an unknown slice name
+ * (listing the valid slices), a config edited since the process closed, and a
+ * replay that diverged. The edge never narrates: the CLI prints the
  * version warning, before the unknown-slice refusal; a replay refusal carries
  * it in its message.
  */

package/src/flows/helpers.ts

@@ -174,7 +174,7 @@ const SHAPE_CRITERIA = [
   "land: every stray is incidental (a needed import, a test fixture, a renamed call site).",
   "refactor-first: the strays are a seam change the task needs; land them first as a behaviour-neutral refactor.",
   "reject: the naive pass solved the task the wrong way; discard it and re-implement, no human; strays a refactor could absorb never count as reject.",
-].join("\n")
+].join(" ")
 
 const readable = (c: Change): boolean =>
   (c.status === "deleted" || c.after !== undefined) &&

package/src/replay/judgeDiffShape.test.ts

@@ -7,6 +7,7 @@ import {
   type JudgeAnswer,
   type StepRequest,
 } from "../flows/index.js"
+import { toRequest } from "../judges/index.js"
 import { globMatches } from "../wire/index.js"
 
 // Lives in replay, which wires `matches: globMatches` (flows tests may not import wire).
@@ -123,6 +124,19 @@ describe("judgeDiffShape", () => {
     expect(q?.criteria).toContain("strays a refactor could absorb never count as reject")
   })
 
+  it("is sendable to jev with exactly its three labels", async () => {
+    const requests = install({ shape: { answer: "land", p: 0.9 } })
+    await judgeDiffShape("shape", { changes: naive, allow: ["src/**"] })
+    const request = requests[0]
+    if (request?.kind !== "judge") throw new Error("expected a judge step")
+    const { questions } = toRequest(request.questions, {})
+    expect(Object.keys((questions["shape"] as { criteria: object }).criteria)).toEqual([
+      "land",
+      "refactor-first",
+      "reject",
+    ])
+  })
+
   it("carries exactly four evidence keys, with stray content before/after and in-allowlist paths only", async () => {
     const requests = install({ shape: { answer: "land", p: 0.9 } })
     await judgeDiffShape("shape", { changes: naive, allow: ["src/**"] })

package/src/workflows/feature/review.test.ts

@@ -3,6 +3,7 @@ import {
   installContext,
   type Change,
   type JudgeAnswer,
+  type JudgeQuestion,
   type ScopeOptions,
   type StepRequest,
 } from "../../flows/index.js"
@@ -18,6 +19,7 @@ import type { ReviewNote } from "../../steering/index.js"
 import * as bundled from "../index.js"
 import type { BuildTailOptions as PublicBuildTailOptions } from "./index.js"
 import { fixtureContext } from "../shared/text.fixture.js"
+import { toRequest } from "../../judges/index.js"
 
 afterEach(() => installContext(undefined))
 
@@ -65,6 +67,7 @@ interface Run {
   readonly prompts: Map<string, string>
   readonly result: ReviewOutcome | "stopped"
   readonly judged: readonly string[]
+  readonly asked: readonly JudgeQuestion[]
   /** `acceptClean` of each `review.await-review` gate, in order. */
   readonly gates: readonly boolean[]
   readonly gateMessages: readonly string[]
@@ -78,7 +81,7 @@ interface Run {
 const drive = async (d: Drive): Promise<Run> => {
   const log: string[] = []
   const prompts = new Map<string, string>()
-  const judged: string[] = []
+  const asked: JudgeQuestion[] = []
   const gates: boolean[] = []
   const gateMessages: string[] = []
   const since: string[] = []
@@ -113,7 +116,7 @@ const drive = async (d: Drive): Promise<Run> => {
       gateMessages.push(request.options.message ?? "")
     }
     if (request.kind === "agent") prompts.set(request.name, request.prompt)
-    if (request.kind === "judge") judged.push(...request.questions.map((q) => q.id))
+    if (request.kind === "judge") asked.push(...request.questions)
   }
   const context = fixtureContext(
     { vars: d.vars ?? {} },
@@ -159,7 +162,17 @@ const drive = async (d: Drive): Promise<Run> => {
     if (!(error instanceof Stop)) throw error
     result = "stopped"
   }
-  return { log, prompts, result, judged, gates, gateMessages, since, scopes }
+  return {
+    log,
+    prompts,
+    result,
+    judged: asked.map((q) => q.id),
+    asked,
+    gates,
+    gateMessages,
+    since,
+    scopes,
+  }
 }
 
 const verdict = (answer: string, p = 0.9): JudgeAnswer => ({ answer, p })
@@ -418,6 +431,15 @@ describe("review verdict routing", () => {
     expect(run.log.filter((n) => n === "review.triage")).toHaveLength(1)
   })
 
+  it("asks only questions sendable to jev, each choice with its labels", async () => {
+    const run = await drive({ notes: ["add — a"] })
+    const { questions } = toRequest(run.asked, {})
+    const labels = (id: string) => Object.keys((questions[id] as { criteria: object }).criteria)
+    expect(labels("note-1")).toEqual(["edit", "question", "nit", "praise"])
+    expect(labels("note-1.reversibility")).toEqual(["one-way", "two-way"])
+    expect(labels("note-1.blastRadius")).toEqual(["local", "module", "contract"])
+  })
+
   it("an unanswered note counts as edit", async () => {
     const run = await drive({ notes: ["add — typo", "sub", "mul"], answers: {} })
     expect(run.log).toEqual([