@pmelab/gtd 55.0.0 → 57.0.0

package/src/workflows/feature/reviewer-walkthrough.feature

@@ -3,16 +3,14 @@ Feature: a reviewer walks the human through the feature's code
 
   The feature's final review is written by a reviewer turn
   (`build.review.reviewing`): chunks grouped by concern, each pointing at its
-  code, so the phone shows it. Every finding the quality laps left gets a chunk
-  of its own, headed exactly as the prompt gives it, holding the pointers to the
-  code it cites; gtd then adds the finding's text and the answer boxes
-  (`build.review.carrying`). A chunk pointing at no code, or a finding left out,
-  is refused.
+  code, so the phone shows it. The quality lap's findings were decided at the
+  finding gate: the reviewer is told nothing of them, and gtd only appends
+  the lap's summary (`build.review.carrying`). A chunk pointing at no code is
+  refused.
 
   Background:
     Given the shipped workflows, as gtd install scaffolds them
-    And an environment variable "GTD_PROTOTYPEROUNDS" set to "0"
-    And an environment variable "GTD_QUALITYREVIEWS" set to "correctness"
+    And an environment variable "GTD_QUALITYREVIEWS" set to ""
     And an environment variable "GTD_FLAGMODE" set to "false"
     And a file "src/calc.ts" with:
       """
@@ -53,30 +51,9 @@ Feature: a reviewer walks the human through the feature's code
     And step "implement.applying" runs its script
     And step "implement.health.check" changes nothing
     And step "implement.closing" runs its script
-    And step "build.quality.correctness.reviewing" writes ".gtd/QUALITY.md" with:
-      """
-      # Quality
-
-      ## AC-1 ignores floating point
-
-      Should subtract round? The criterion does not say.
-      """
-    And judge step "build.quality.triage" answers:
-      """
-      [{"id":"Q1","answer":"design","p":0.9}]
-      """
-    And step "build.quality.sorting" runs its script
-    And step "build.quality.closing" runs its script
-    And step "build.review.carrying" runs its script
-    And judge step "build.review.ranking" answers:
-      """
-      []
-      """
 
-  Scenario: the finding gets its text and answer boxes under the reviewer's pointers
-    Given step "build.review.reviewing" is prompted with "- `## AC-1 ignores floating point`:"
-    And step "build.review.reviewing" is prompted with "> Should subtract round? The criterion does not say."
-    And step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
+  Scenario: the review is the reviewer's walkthrough, as it wrote it
+    Given step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
       """
       # Review: abc1234
 
@@ -87,47 +64,24 @@ Feature: a reviewer walks the human through the feature's code
       The new export.
 
       - [ ] ./src/calc.ts#2-2 — subtract(a, b)
-
-      ## AC-1 ignores floating point
-
-      - [ ] ./src/calc.ts#2-2 — returns the raw difference
       """
     When the process is driven
     Then the process rests at "build.review.await-review"
+    And the resting step's content does not contain "Tick exactly one"
     And ".gtd/REVIEW.md" contains:
       """
-      ## Findings
-
-      ### AC-1 ignores floating point
-
-      _critical: judge unsure_
-
-      Should subtract round? The criterion does not say.
-
-      In the code:
-
-      - [ ] ./src/calc.ts#2-2 — returns the raw difference
-
-      Tick exactly one:
-
-      - [ ] fix as suggested
-      - [ ] won't fix
-      - [ ] _your answer_
-
-      ## Walkthrough
+      <!-- base: 0000000 -->
 
-      ### Subtract
+      ## Subtract
 
       The new export.
 
       - [ ] ./src/calc.ts#2-2 — subtract(a, b)
       """
-    And no commit body contains "build.review.composing"
     And the commit subjects end with:
       """
-      gtd(agent): build.review.reviewing → build.review.ranking
-      gtd(judge): build.review.ranking → build.review.carrying
-      gtd(check): build.review.carrying → build.review.await-review
+      gtd(check): implement.closing → build.review.reviewing
+      gtd(agent): build.review.reviewing → build.review.await-review
       """
 
   Scenario: a chunk pointing at no code is refused
@@ -140,10 +94,6 @@ Feature: a reviewer walks the human through the feature's code
       ## Subtract
 
       The new export, with no pointer.
-
-      ## AC-1 ignores floating point
-
-      - [ ] ./src/calc.ts#2-2 — returns the raw difference
       """
     Then driving the process is refused with "Subtract — points at no code"
 
@@ -158,26 +108,9 @@ Feature: a reviewer walks the human through the feature's code
 
       ### Subtract
 
-      - [ ] ./src/calc.ts#2-2 — subtract(a, b)
-
-      ### AC-1 ignores floating point
-
-      - [ ] ./src/calc.ts#2-2 — returns the raw difference
-      """
-    Then driving the process is refused with "Walkthrough — gtd sorts the chunks into sections; write `## ` chunks only"
-
-  Scenario: a finding the reviewer left out is refused
-    Given step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
-      """
-      # Review: abc1234
-
-      <!-- base: 0000000 -->
-
-      ## Subtract
-
       - [ ] ./src/calc.ts#2-2 — subtract(a, b)
       """
-    Then driving the process is refused with "AC-1 ignores floating point — no chunk of its own"
+    Then driving the process is refused with "Walkthrough — write `## ` chunks only"
 
   Scenario: a standard conflict the reviewer writes itself is refused
     Given step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
@@ -189,31 +122,10 @@ Feature: a reviewer walks the human through the feature's code
       ## Standard conflict 0123abcd
 
       - [ ] ./src/calc.ts#2-2 — copied over
-
-      ## AC-1 ignores floating point
-
-      - [ ] ./src/calc.ts#2-2 — returns the raw difference
       """
     Then driving the process is refused with "Standard conflict 0123abcd — gtd writes conflicts"
 
   Scenario: two chunks with one title are refused
-    Given step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
-      """
-      # Review: abc1234
-
-      <!-- base: 0000000 -->
-
-      ## AC-1 ignores floating point
-
-      - [ ] ./src/calc.ts#2-2 — subtract(a, b)
-
-      ## AC-1 ignores floating point
-
-      - [ ] ./src/calc.ts#2-2 — returns the raw difference
-      """
-    Then driving the process is refused with "AC-1 ignores floating point — a second chunk with this title"
-
-  Scenario: a risk marked in a finding's chunk is refused, never fixed
     Given step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
       """
       # Review: abc1234
@@ -224,8 +136,8 @@ Feature: a reviewer walks the human through the feature's code
 
       - [ ] ./src/calc.ts#2-2 — subtract(a, b)
 
-      ## AC-1 ignores floating point
+      ## Subtract
 
-      - [ ] ./src/calc.ts#2-2 — Risk: no rounding
+      - [ ] ./src/calc.ts#2-2 — returns the raw difference
       """
-    Then driving the process is refused with "no `Risk:` in a quality finding's chunk"
+    Then driving the process is refused with "Subtract — a second chunk with this title"

package/src/workflows/feature/seam-rejection.feature

@@ -9,7 +9,6 @@ Feature: rejected naive passes walk one fixed path through the steer gate to the
 
   Background:
     Given the shipped workflows, as gtd install scaffolds them
-    And an environment variable "GTD_PROTOTYPEROUNDS" set to "0"
     And an environment variable "GTD_QUALITYREVIEWS" set to ""
     And an environment variable "GTD_FLAG_RISK_TIER" set to "0"
     And a file "src/calc.ts" with:

package/src/workflows/feature/seam.feature

@@ -7,7 +7,6 @@ Feature: the naive pass and seam judge decide how a feature is built
 
   Background:
     Given the shipped workflows, as gtd install scaffolds them
-    And an environment variable "GTD_PROTOTYPEROUNDS" set to "0"
     And the gtd config ends with:
       """
       export const config = {

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

@@ -403,8 +403,7 @@ describe("implement", () => {
     lastOptions = undefined
     installContext(
       fixtureContext(
-        // The hands-on test has its own scenarios.
-        { vars: { prototypeRounds: "0" } },
+        {},
         {
           step: (request: StepRequest) => {
             requests.push(request)

package/src/workflows/feature/seam.ts

@@ -3,6 +3,7 @@ import {
   agent,
   applyScript,
   changes,
+  changesSince,
   check,
   codeChanges,
   env,
@@ -14,6 +15,7 @@ import {
   read,
   refuse,
   removeScript,
+  restoreScript,
   run,
   scope,
   vars,
@@ -22,11 +24,10 @@ import {
   type DiffShapeResult,
   type KeptTip,
 } from "../../flows/index.js"
-import { coder, FEEDBACK, PROTOTYPE, REFACTOR_NOTE, REQUIREMENTS } from "../shared/steps.js"
+import { coder, FEEDBACK, NAIVE_TEST, REFACTOR_NOTE, REQUIREMENTS } from "../shared/steps.js"
 import * as t from "../shared/text.js"
 import { defaults } from "../shared/vars.js"
 import { green } from "./green.js"
-import { prototypeTest } from "./prototype.js"
 import { criterionIds, requireCriteria, requireCriteriaFrozen } from "./criteria.js"
 import {
   COMMENT,
@@ -36,7 +37,7 @@ import {
   reviewedRefactor,
 } from "./refactor.js"
 import { isCode } from "../shared/review.js"
-import { flagModeOn } from "../shared/options.js"
+import { flagModeOn, naiveTestOn } from "../shared/options.js"
 import type { RiskTier } from "./tier.js"
 
 /** Naive-pass retries after a `reject` or a red suite; the next one rests at `seam.steer`. */
@@ -53,7 +54,7 @@ export const judgeSeam = (tip: KeptTip): Promise<DiffShapeResult> =>
     exempt: tip.changes.filter((c) => c.status === "added").map((c) => c.path),
   })
 
-export const naivePass = (retry?: t.NaiveRetry): Promise<void> =>
+export const naivePass = (retry?: t.NaiveRetry | t.NaiveTested): Promise<void> =>
   t.agentWithSkills("naive", t.naivePassPrompt(retry), {
     label: retry === undefined ? "Writing the naive pass" : "Rewriting the naive pass",
     keep: true,
@@ -226,9 +227,8 @@ export const implementing = (options: ImplementOptions): Promise<readonly string
     if (built) await build(options)
     else requireCriteriaFrozen(REQUIREMENTS)
     await green()
-    await prototypeTest()
     const ids = criterionIds(read(REQUIREMENTS) ?? "")
-    await run("closing", removeScript([REQUIREMENTS, REFACTOR_NOTE, PROTOTYPE]), {
+    await run("closing", removeScript([REQUIREMENTS, REFACTOR_NOTE]), {
       label: "Closing out the implementation",
     })
     return ids
@@ -277,7 +277,7 @@ const budget = (): ((next: t.NaiveRetry) => Promise<Verdict>) => {
 }
 
 interface Vetting {
-  readonly again: ReturnType<typeof budget>
+  again: ReturnType<typeof budget>
   /** Set once a human overrules a test fault: the rule does not fit this change, so later passes skip it. */
   waived: boolean
 }
@@ -304,16 +304,62 @@ const ruled = async (tip: KeptTip, again: Vetting["again"]): Promise<Decision |
   return next.kind === "overruled" ? { reference: tip.hash, tip } : next
 }
 
+/** Back to `before`: every code path, and the notes file. `.gtd/` edits (the requirements) stay. */
+const untryScript = (before: string): string => {
+  const code = changesSince(before).filter((c) => isCode(c.path))
+  return restoreScript(
+    head(),
+    {
+      restore: code.filter((c) => c.status !== "added").map((c) => c.path),
+      remove: [...code.filter((c) => c.status === "added").map((c) => c.path), NAIVE_TEST],
+    },
+    before,
+  )
+}
+
+/**
+ * A person tries the green pass `tip` in the working tree. Resolves
+ * `undefined` on a clean landing (signed off), else what they sent back;
+ * either way the pass, their code edits and the notes file leave the tree.
+ */
+const tried = async (tip: KeptTip): Promise<t.NaiveTested | undefined> => {
+  const before = head()
+  await run("applying", applyScript(tip.hash, FEEDBACK, t.naiveApplyFailure(tip.hash), [".gtd"]), {
+    label: "Applying the naive pass to try it",
+  })
+  await human("try", {
+    message: t.naiveTryMessage(flagModeOn()),
+    label: "Try the naive pass",
+    file: NAIVE_TEST,
+    acceptClean: true,
+  })
+  const landed = changes()
+  const edits = landed.some((c) => isCode(c.path)) ? head() : undefined
+  const notes = read(NAIVE_TEST)?.trim() || undefined
+  if (landed.some((c) => c.path === REQUIREMENTS)) requireCriteria(REQUIREMENTS)
+  await run("restoring", untryScript(before), { label: "Taking the naive pass out again" })
+  return landed.length === 0 ? undefined : { tested: tip.hash, notes, edits }
+}
+
 const decide = (): Promise<Decision> =>
   scope("seam", async () => {
     const vetting: Vetting = { again: budget(), waived: false }
-    let retry: t.NaiveRetry | undefined
+    let retry: t.NaiveRetry | t.NaiveTested | undefined
     for (;;) {
       await naivePass(retry)
       const tip = kept("naive")!
+      const verdict = await vetted(tip, vetting)
+      if (verdict === undefined && naiveTestOn()) {
+        const sentBack = await tried(tip)
+        if (sentBack !== undefined) {
+          // A person's feedback is no failed attempt: the next pass gets a full retry budget.
+          vetting.again = budget()
+          retry = sentBack
+          continue
+        }
+      }
       const outcome =
-        (await vetted(tip, vetting)) ??
-        (flagModeOn() ? await ruled(tip, vetting.again) : { reference: tip.hash, tip })
+        verdict ?? (flagModeOn() ? await ruled(tip, vetting.again) : { reference: tip.hash, tip })
       if (!("kind" in outcome)) return outcome
       retry = outcome.kind === "retry" ? outcome.retry : undefined
     }

package/src/workflows/feature/standards.feature

@@ -1,8 +1,8 @@
 @inmem
 Feature: the review learns standards from the human's answers
 
-  A won't-fix tick or an own answer at the review gate is a candidate
-  standard. A judge asks whether it is reusable; above `standardMinP` it is
+  A won't-fix tick or an own answer at the review gate — on a note the human
+  added with its own answer boxes — is a candidate standard. A judge asks whether it is reusable; above `standardMinP` it is
   kept in `.gtd/STANDARDS.md` with no further question, and every later agent
   turn — lenses and coder alike — is told to follow it. One that contradicts
   a standard already kept becomes a note in `.gtd/REVIEW.md` the review cannot
@@ -11,8 +11,7 @@ Feature: the review learns standards from the human's answers
 
   Background:
     Given the shipped workflows, as gtd install scaffolds them
-    And an environment variable "GTD_PROTOTYPEROUNDS" set to "0"
-    And an environment variable "GTD_QUALITYREVIEWS" set to "correctness"
+    And an environment variable "GTD_QUALITYREVIEWS" set to ""
     And an environment variable "GTD_FLAGMODE" set to "false"
     And a file "src/calc.ts" with:
       """
@@ -59,15 +58,6 @@ Feature: the review learns standards from the human's answers
     And step "implement.applying" runs its script
     And step "implement.health.check" changes nothing
     And step "implement.closing" runs its script
-    And step "build.quality.correctness.reviewing" writes ".gtd/QUALITY.md" with:
-      """
-      # Quality
-
-      ## subtract has no class
-
-      Calculators read better as classes.
-      """
-    And step "build.quality.sorting" runs its script
     And step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
       """
       # Review: abc1234
@@ -78,29 +68,7 @@ Feature: the review learns standards from the human's answers
 
       - [ ] ./src/calc.ts#2-2 — subtract(a, b)
       """
-    And step "build.review.reviewing", when prompted with "- `## subtract has no class`", writes ".gtd/REVIEW.md" with:
-      """
-      # Review: abc1234
-
-      <!-- base: 0000000 -->
-
-      ## Subtract
-
-      - [ ] ./src/calc.ts#2-2 — subtract(a, b)
-
-      ## subtract has no class
-
-      - [ ] ./src/calc.ts#2-2 — a plain function
-      """
     And step "build.review.carrying" runs its script
-    And judge step "build.review.ranking" answers:
-      """
-      []
-      """
-    And judge step "build.quality.triage" answers:
-      """
-      [{"id":"Q1","answer":"design","p":0.9}]
-      """
     And step "build.review.closing" runs its script
     And step "build.review.standards.recording" runs its script
     And step "build.review.standards.raising" runs its script
@@ -111,6 +79,18 @@ Feature: the review learns standards from the human's answers
       [{"id":"standard-1","answer":"yes","p":0.9}]
       """
     When the process is driven
+    And ".gtd/REVIEW.md" gets appended:
+      """
+      ## subtract has no class
+
+      Calculators read better as classes.
+
+      Tick exactly one:
+
+      - [ ] fix as suggested
+      - [ ] won't fix
+      - [ ] _your answer_
+      """
     And the text "- [ ] won't fix" in ".gtd/REVIEW.md" is replaced with "- [x] won't fix"
     And gtd lands
     And the process is driven
@@ -143,6 +123,18 @@ Feature: the review learns standards from the human's answers
       """
     And step "build.health.check" changes nothing
     When the process is driven
+    And ".gtd/REVIEW.md" gets appended:
+      """
+      ## subtract has no class
+
+      Calculators read better as classes.
+
+      Tick exactly one:
+
+      - [ ] fix as suggested
+      - [ ] won't fix
+      - [ ] _your answer_
+      """
     And the text "- [ ] _your answer_" in ".gtd/REVIEW.md" is replaced with "- [x] Calculators are plain functions."
     And gtd lands
     And the process is driven
@@ -162,6 +154,18 @@ Feature: the review learns standards from the human's answers
       [{"id":"standard-1","answer":"yes","p":0.6}]
       """
     When the process is driven
+    And ".gtd/REVIEW.md" gets appended:
+      """
+      ## subtract has no class
+
+      Calculators read better as classes.
+
+      Tick exactly one:
+
+      - [ ] fix as suggested
+      - [ ] won't fix
+      - [ ] _your answer_
+      """
     And the text "- [ ] won't fix" in ".gtd/REVIEW.md" is replaced with "- [x] won't fix"
     And gtd lands
     And the process is driven
@@ -210,6 +214,18 @@ Feature: the review learns standards from the human's answers
       [{"id":"answer-1","answer":"praise","p":0.9}]
       """
     When the process is driven
+    And ".gtd/REVIEW.md" gets appended:
+      """
+      ## subtract has no class
+
+      Calculators read better as classes.
+
+      Tick exactly one:
+
+      - [ ] fix as suggested
+      - [ ] won't fix
+      - [ ] _your answer_
+      """
     And the text "- [ ] _your answer_" in ".gtd/REVIEW.md" is replaced with "- [x] Calculators are plain functions."
     And gtd lands
     And the process is driven
@@ -267,6 +283,18 @@ Feature: the review learns standards from the human's answers
       [{"id":"standard-1","answer":"yes","p":0.9},{"id":"standard-1.contradicts-1","answer":"yes","p":0.8}]
       """
     When the process is driven
+    And ".gtd/REVIEW.md" gets appended:
+      """
+      ## subtract has no class
+
+      Calculators read better as classes.
+
+      Tick exactly one:
+
+      - [ ] fix as suggested
+      - [ ] won't fix
+      - [ ] _your answer_
+      """
     And the text "- [ ] won't fix" in ".gtd/REVIEW.md" is replaced with "- [x] won't fix"
     And gtd lands
     And the process is driven
@@ -303,6 +331,18 @@ Feature: the review learns standards from the human's answers
       [{"id":"standard-1","answer":"yes","p":0.9},{"id":"standard-1.contradicts-1","answer":"yes","p":0.8}]
       """
     When the process is driven
+    And ".gtd/REVIEW.md" gets appended:
+      """
+      ## subtract has no class
+
+      Calculators read better as classes.
+
+      Tick exactly one:
+
+      - [ ] fix as suggested
+      - [ ] won't fix
+      - [ ] _your answer_
+      """
     And the text "- [ ] won't fix" in ".gtd/REVIEW.md" is replaced with "- [x] won't fix"
     And gtd lands
     And the process is driven

package/src/workflows/fix/fix-lens-origin.feature

@@ -1,9 +1,8 @@
 @inmem
-Feature: each finding the reviewer is handed names the lens that wrote it
+Feature: each finding at the finding gate names the lens that wrote it
 
   Lenses append to `.gtd/QUALITY.md` in turn, so a finding belongs to the lens
-  whose turn added it — even when two lenses give findings the same title,
-  which the review then tells apart by a numbered suffix.
+  whose turn added it — even when two lenses give findings the same title.
 
   Background:
     Given the shipped workflows, as gtd install scaffolds them
@@ -62,7 +61,7 @@ Feature: each finding the reviewer is handed names the lens that wrote it
 
       From tests.
       """
-    And step "build.quality.sorting" runs its script
+    And step "build.quality.gating" runs its script
     And step "build.quality.closing" runs its script
     And step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
       """
@@ -84,8 +83,8 @@ Feature: each finding the reviewer is handed names the lens that wrote it
       """
     When gtd opens door "fix"
     And the process is driven
-    Then the process rests at "build.review.await-review"
-    And ".gtd/REVIEW.md" contains:
+    Then the process rests at "build.quality.gate"
+    And ".gtd/QUALITY.md" contains:
       """
       ## sum is untested
 
@@ -93,9 +92,23 @@ Feature: each finding the reviewer is handed names the lens that wrote it
 
       From correctness.
 
-      ## sum is untested (2)
+      - [ ] fix
+
+      ## sum is untested
 
       _From lens tests._
 
       From tests.
+
+      - [ ] fix
+      """
+    When gtd lands
+    And the process is driven
+    Then the process rests at "build.review.await-review"
+    And ".gtd/REVIEW.md" contains:
+      """
+      Dropped at the finding gate:
+
+      - sum is untested (lens correctness)
+      - sum is untested (lens tests)
       """