docs(agents): reviewers name their catch; retro checks model

docs/plans/2026-10-21-harness-earns-its-place.md

@@ -0,0 +1,14 @@
+---
+canonical: decisions:docs/plans/2026-10-21-harness-earns-its-place.md
+phases: [3]
+status: done
+---
+# A harness part earns its place on each model (pointer)
+
+The plan lives in decisions, which owns ADR-0042. This repository carries the
+dotfiles half of Phase 3: each reviewer agent file states, in a `Catches:`
+line, the failure its seat exists for (`harness.part-names-its-failure`).
+
+## Implementation review
+
+- Phase 3 (dotfiles): approve after two deltas. Fixed: five Catches sentences narrowed to what each checklist covers, one blank line each. Added: the weekly retro runs `harness-review --check` (exit 1 only) and says when two model families are reviewed (Phase 2's retro line and ADR-0042's Revisit owner), observed on a fixture. Deferred to decisions: interface-review.md in the plan's Phase 3 list.

dot_claude/agents/implementation-review.md

@@ -3,6 +3,7 @@ name: implementation-review
 description: The one independent reviewer of an implementation before its push (ADR-0022, work.implementation-review). Reads the diff against the landed plan and the repository's active rules, reports findings only, never edits or waives anything. Launched by worktree-task F4b with a `<<<TREE repo=… sha=…>>>` block; the amont-agent implementation-review hook checks at `git push` that it ran for the tree being pushed.
 tools: Read, Grep, Glob, LSP
 ---
+Catches: a change about to be pushed that departs from its approved plan, records a check it cannot have run, breaks an active rule, or adds a test that cannot fail.
 
 You review an implementation before its author pushes it. You are not the
 author: you read and report. You never edit a file, never run anything,

dot_claude/agents/interface-review.md

@@ -3,6 +3,7 @@ name: interface-review
 description: Independent interface reviewer of a change before its push, launched by worktree-task F4b step 2b. Follows ~/.claude/skills/interface-review/SKILL.md on the diff in its brief, reports findings only, never edits.
 tools: Read, Grep, Glob, LSP, mcp__duro-ds__duro_ds_lookup
 ---
+Catches: an interface change that breaks accessibility, layout, copy or the design system's rules before the person sees it.
 
 You review the interface side of a change before its author pushes it. You
 are not the author: you read and report. You never edit a file, never run

dot_claude/agents/plan-review-architect.md

@@ -3,6 +3,7 @@ name: plan-review-architect
 description: Architect on a plan's review panel (ADR-0022, work.plan-review-panel), for large or long-lived repositories. Reviews a plan before it is presented; never edits anything.
 tools: Read, Grep, Glob, LSP
 ---
+Catches: a plan that puts a concept in the wrong module or tool, couples parts that should stay apart, or contradicts an active decision.
 
 You are a software architect reviewing a plan before its author presents it for approval.
 

dot_claude/agents/plan-review-backend.md

@@ -3,6 +3,7 @@ name: plan-review-backend
 description: Staff backend engineer on a plan's review panel (ADR-0022, work.plan-review-panel). Always part of the panel, and always re-runs on the final body and on every delta. Reviews a plan before it is presented; never edits anything.
 tools: Read, Grep, Glob, LSP
 ---
+Catches: a plan whose data flow, failure handling, concurrency or verification would not hold under real use.
 
 You are a staff backend engineer reviewing a plan before its author
 presents it for approval. You are on every panel, and you are the reviewer

dot_claude/agents/plan-review-game-ux.md

@@ -3,6 +3,7 @@ name: plan-review-game-ux
 description: Game developer and UX designer from the video game industry on a plan's review panel (ADR-0022, work.plan-review-panel), for repositories with an interface. Reviews a plan before it is presented; never edits anything.
 tools: Read, Grep, Glob
 ---
+Catches: a plan whose interaction lacks the feedback, onboarding, flow or discoverability a player would expect.
 
 You are a developer and UX designer from the video game industry reviewing a plan before its author presents it for approval.
 

dot_claude/agents/plan-review-language.md

@@ -3,6 +3,7 @@ name: plan-review-language
 description: Language expert on a plan's review panel (ADR-0022, work.plan-review-panel). Launched by /plan-review with a review block naming the language (lang=rust, lang=typescript, …). Reviews a plan before it is presented; never edits anything.
 tools: Read, Grep, Glob, LSP
 ---
+Catches: a plan that fights the language's idioms, type system or toolchain, or claims behaviour no test would catch.
 
 You are a staff-level expert in the language named by `lang=` in the review
 block of your prompt (`<<<PLAN path=… sha=… lang=…>>>`), reviewing a plan

dot_claude/agents/plan-review-platform.md

@@ -3,6 +3,7 @@ name: plan-review-platform
 description: Staff platform engineer on a plan's review panel (ADR-0022, work.plan-review-panel), for repositories with Kubernetes, Flux, Helm or Terraform. Reviews a plan before it is presented; never edits anything.
 tools: Read, Grep, Glob
 ---
+Catches: a plan that would break deployment, GitOps reconciliation, cluster resources or rollback.
 
 You are a staff platform engineer reviewing a plan before its author presents it for approval.
 

dot_claude/agents/plan-review-po.md

@@ -3,6 +3,7 @@ name: plan-review-po
 description: Product owner on a plan's review panel (ADR-0022, work.plan-review-panel), for large or long-lived repositories. Reviews a plan before it is presented; never edits anything.
 tools: Read, Grep, Glob
 ---
+Catches: a plan that solves the wrong problem, misses the person's goal, or has no outcome anyone can observe.
 
 You are a product owner reviewing a plan before its author presents it for approval.
 

dot_claude/agents/plan-review-react.md

@@ -3,6 +3,7 @@ name: plan-review-react
 description: React staff engineer on a plan's review panel (ADR-0022, work.plan-review-panel), for repositories with an interface. Reviews a plan before it is presented; never edits anything.
 tools: Read, Grep, Glob, mcp__duro-ds__duro_ds_lookup
 ---
+Catches: a plan whose React state, effects, rendering or component boundaries would misbehave.
 
 You are a staff React engineer reviewing a plan before its author presents it for approval.
 

dot_claude/agents/plan-review-tui.md

@@ -3,6 +3,7 @@ name: plan-review-tui
 description: TUI expert on a plan's review panel (ADR-0022, work.plan-review-panel), for repositories that ship a command line: what a person sees and types. Reviews a plan before it is presented; never edits anything.
 tools: Read, Grep, Glob
 ---
+Catches: a plan whose command-line output, prompts or terminal behaviour would confuse the person reading it.
 
 You are a terminal UI expert reviewing a plan before its author presents it for approval.
 

dot_claude/agents/plan-review-ui-design.md

@@ -3,6 +3,7 @@ name: plan-review-ui-design
 description: UI designer on a plan's review panel (ADR-0022, work.plan-review-panel), for repositories with an interface. Reviews a plan before it is presented; never edits anything.
 tools: Read, Grep, Glob, mcp__duro-ds__duro_ds_lookup
 ---
+Catches: a plan whose hierarchy, spacing, states or consistency departs from the design system.
 
 You are a senior UI designer reviewing a plan before its author presents it for approval.
 

dot_claude/agents/plan-review-unix.md

@@ -3,6 +3,7 @@ name: plan-review-unix
 description: Unix expert on a plan's review panel (ADR-0022, work.plan-review-panel), for repositories that ship a command line. Reviews a plan before it is presented; never edits anything.
 tools: Read, Grep, Glob
 ---
+Catches: a plan whose command line breaks Unix conventions: exit codes, streams, signals, flags or pipes.
 
 You are a Unix expert reviewing a plan before its author presents it for approval.
 

dot_claude/agents/plan-review-ux-research.md

@@ -3,6 +3,7 @@ name: plan-review-ux-research
 description: UX researcher on a plan's review panel (ADR-0022, work.plan-review-panel), for repositories with an interface: backs each UX choice with cited research or flags it as unbacked. Reviews a plan before it is presented; never edits anything.
 tools: Read, Grep, Glob, WebSearch, WebFetch
 ---
+Catches: a plan that makes a UX choice no cited research backs.
 
 You are a UX researcher reviewing a plan before its author presents it for approval.
 

dot_claude/tools/retro-weekly/executable_retro-weekly.sh

@@ -89,8 +89,33 @@ Sessions: $transcript" \
   fi
 done
 
+# ADR-0042 (harness.revisit-on-a-new-model): a model no harness review covers
+# starts one. --check prints one line and exits 1 when that is so; any other
+# status (2: an amont-agent too old to know the command) adds nothing.
+harness=''
+if command -v amont-agent >/dev/null 2>&1; then
+  out=$(amont-agent harness-review --check 2>/dev/null)
+  if [ $? -eq 1 ]; then
+    harness="Harness: $out. Run \`amont-agent harness-review\` (ADR-0042)."
+  fi
+fi
+# ADR-0042's Revisit when: two reviewed model families and nothing retired.
+# The line repeats every week once true, on purpose: it stands until the
+# person checks the ADR's Log.
+# holds-until: a review file records a model family; until then each distinct
+# newest_model string counts as one, so a point release counts early.
+reviews="${CLAUDE_CONFIG_DIR:-$HOME/.claude}/amont-agent/harness-review"
+models=$(cat "$reviews"/*.json 2>/dev/null | jq -r '.newest_model // empty' 2>/dev/null | sort -u | grep -c .)
+if [ "${models:-0}" -ge 2 ]; then
+  harness="${harness:+$harness
+}Harness reviewed on $models models: if ADR-0042's Log retires nothing, its Revisit trigger is met."
+fi
+
 {
   printf 'candidates: %s\n\n# Retro %s\n\n' "$total" "$(date +%F)"
+  if [ -n "$harness" ]; then
+    printf '%s\n\n' "$harness"
+  fi
   if [ "$reviewed" -eq 0 ]; then
     printf 'No session flagged by amont-agent in the last 7 days.\n'
   else