evals/: port Modern AI Productivity Pack Component 4, wire fm-brief.sh eval gate, Brandon-voice tov-eval

.claude/settings.json

@@ -51,5 +51,6 @@
         ]
       }
     ]
-  }
+  },
+  "autoCompactEnabled": true
 }

.pi/extensions/fm-calm.ts

@@ -1,6 +1,6 @@
 // Firstmate's home-persistent Pi transcript presentation toggle.
 //
-// Compatibility boundary: Pi 0.81.1 and 0.82.0 expose built-in ToolDefinitions, per-slot
+// Compatibility boundary: Pi 0.81.1 through 0.82.1 expose built-in ToolDefinitions, per-slot
 // renderers, renderShell: "self", session_start replacement reasons,
 // ExtensionUIContext.setToolsExpanded(), setWorkingVisible(), and
 // setHiddenThinkingLabel(). The focused tests pin those assumptions. Version-bounded

.pi/extensions/lib/fm-calm-operational-user-layout.ts

@@ -1,4 +1,4 @@
-// Pi 0.81.1 and 0.82.0 add the ordinary-user spacer and row together.
+// Pi 0.81.1 through 0.82.1 add the ordinary-user spacer and row together.
 // This version-bounded adapter changes only that presentation and never message delivery.
 import {
   InteractiveMode,

AGENTS.md

@@ -11,6 +11,11 @@ Use light nautical seasoning only when it fits: the occasional "aye", "on deck",
 Keep that seasoning optional and never let it obscure technical content; never use it in commits, briefs, PRs, or anything crewmates or other tools read; drop the playful flavor entirely when delivering bad news or relaying serious findings.
 For captain-facing escalation style and outcome phrasing, see section 9.
 
+Voice is opted in for this home and carries the interaction, not chat text (captain's explicit preference, 2026-07-19, reinforced 2026-07-20 with "I need to hear you or else I won't use you that much" - take that at face value, this is a retention-critical preference, not a nice-to-have): the captain should know what's going on from the spoken line alone, and only read chat for detail, a reminder, or a reference.
+Run Alfred's global NATIVE-mode voice protocol (`~/.claude/CLAUDE.md`, acknowledge / update / conclude) on every response even while operating as first mate. Bias toward speaking: fire a spoken line, written like a standalone CONCLUDE (self-sufficient, no "read the text for context"), on every turn that carries any real content - a wake handled, a PR opened, a merge landed, a worker dispatched or blocked, a decision surfaced, a finding, a plan, an answer to a question. Only skip the voice call on a genuinely contentless re-arm cycle (an idle-pane flag with zero news, back to back with the last one) - when in doubt, speak.
+Every spoken line still passes through this file's own rules first: address the captain as above, and translate through section 9's outcome language before speaking it, exactly as if it were chat text.
+Never speak an internal term section 9 would forbid in chat.
+
 ## 1. Identity and prime directives
 
 You are the captain's only point of contact for all software work across all of their projects.
@@ -37,7 +42,7 @@ Hard rules, in priority order:
    If work failed, say so plainly with the evidence.
 
 You may maintain this repo's private operational state directly.
-Shared tracked material is `AGENTS.md`, `README.md`, `CONTRIBUTING.md`, `.tasks.toml`, `.github/workflows/`, `bin/`, `.agents/skills/`, and public `skills/`.
+Shared tracked material is `AGENTS.md`, `README.md`, `CONTRIBUTING.md`, `.tasks.toml`, `.github/workflows/`, `bin/`, `.agents/skills/`, `evals/`, and public `skills/`.
 When any crewmate is live, delegate changes to shared tracked material rather than competing with supervision; when the fleet is empty, firstmate may change it directly.
 This repo is a shared template, while `.env`, `data/`, `state/`, `config/`, `projects/`, and `.no-mistakes/` are captain-private and gitignored.
 Ship shared tracked changes through this repo's no-mistakes pipeline and PR path, with the same merge authority as any other project.
@@ -62,6 +67,7 @@ README.md            public overview and development notes
 .claude/skills       symlink to .agents/skills for claude compatibility
 skills/              standalone public installer-facing skills, committed; not loaded by firstmate
 bin/                 helper scripts, committed; read each script's header before first use
+evals/               shared crewmate self-grading checks, committed; ported from the Modern AI Productivity Pack (resources/modern-ai-productivity-pack/03-evals/ in the command-center repo); see evals/README.md; run_eval.sh works standalone and bin/fm-brief.sh's ship-mode scaffold wires the relevant eval(s) into the generated Definition of done
 .env                 optional X-mode pairing token; LOCAL, gitignored; presence-gates section 14
 config/crew-harness  crewmate harness override; LOCAL, gitignored; absent or "default" = same as firstmate. Inherited as the literal file: a concrete primary adapter value also controls a secondmate home's own crewmates (section 4)
 config/crew-dispatch.json  optional crewmate dispatch profiles; LOCAL, gitignored; firstmate-maintained but human-editable natural-language rules that choose a per-task harness/model/effort profile (section 4). Inherited by secondmate homes

CONTRIBUTING.md

@@ -35,7 +35,7 @@ See the [no-mistakes quick start](https://kunchenguid.github.io/no-mistakes/star
 
 - This repo is a template for running a firstmate orchestrator agent.
   `AGENTS.md` is the agent's main job description and names when to load bundled firstmate skills; `CLAUDE.md` is a symlink to it, and `.claude/skills` is a symlink to `.agents/skills`.
-- Only shared material is tracked: `AGENTS.md`, `README.md`, `CONTRIBUTING.md`, `.tasks.toml`, `.github/workflows/`, `bin/`, `.agents/skills/`, and `skills/`.
+- Only shared material is tracked: `AGENTS.md`, `README.md`, `CONTRIBUTING.md`, `.tasks.toml`, `.github/workflows/`, `bin/`, `.agents/skills/`, `evals/`, and `skills/`.
   `.agents/skills/` holds agent-loaded skills that assume a live firstmate home and carry `metadata.internal: true` so installers such as [skills.sh](https://skills.sh) hide them from discovery; `skills/` holds standalone, installer-facing public skills with no firstmate dependency (see the README's "Two-tier skill layout").
   Everything personal to one captain's fleet (`.env`, `data/`, `state/`, `config/`, `projects/`, `.no-mistakes/`) is gitignored; never commit it.
   The root `.tasks.toml` is tracked `tasks-axi` config for `data/backlog.md`; compatible `tasks-axi` is the default backend for routine backlog mutations, with the compatibility definition owned by [`docs/configuration.md`](docs/configuration.md) ("Backlog backend").
@@ -56,7 +56,7 @@ See the [no-mistakes quick start](https://kunchenguid.github.io/no-mistakes/star
 
 ## Development
 
-Tracked changes to firstmate itself - `AGENTS.md`, `README.md`, `CONTRIBUTING.md`, `.tasks.toml`, `.github/workflows/`, `bin/`, `.agents/skills/`, and `skills/` - ship through the `no-mistakes` pipeline on a feature branch and require an explicit merge approval.
+Tracked changes to firstmate itself - `AGENTS.md`, `README.md`, `CONTRIBUTING.md`, `.tasks.toml`, `.github/workflows/`, `bin/`, `.agents/skills/`, `evals/`, and `skills/` - ship through the `no-mistakes` pipeline on a feature branch and require an explicit merge approval.
 Before making any such change, load the agent-only `firstmate-coding-guidelines` skill (`.agents/skills/firstmate-coding-guidelines/SKILL.md`).
 It has the knowledge-placement rules that keep `AGENTS.md` from regrowing after each diet pass.
 There is no reliable way for `bin/fm-brief.sh`'s scaffold to detect that a task's repo is firstmate itself, so firstmate adds this skill's load line to firstmate-repo briefs by hand.

Plans/bubbly-stargazing-quilt.md

@@ -0,0 +1,45 @@
+# Merge firstmate's operating discipline into Alfred (global)
+
+## Status: COMPLETE (verified 2026-07-27)
+
+All four items shipped, confirmed by directly reading the live files rather than trusting memory:
+- `~/.claude/SOUL.md` #9 — captain-facing translation rule, broadened to cover both delegated work and Alfred's own tool use.
+- `~/.claude/SOUL.md` #10 — "Investigation is evidence, not authorization" (diagnostic-reasoning discipline).
+- `~/.claude/SOUL.md` #11 — "Never let an open decision go silent" (decision-hold-lifecycle).
+- `~/.claude/CLAUDE.md` — "KNOWLEDGE-PLACEMENT DISCIPLINE" rule, correctly cross-referencing SOUL.md rather than duplicating the behavioral commitments.
+
+## Context
+
+Brandon has been running this session with me acting *as* firstmate — a separate persona/repo (`ops/firstmate/`, loaded from its own `AGENTS.md`) that orchestrates a fleet of crewmate agents with heavy operational discipline: supervision loops, receipt-on-done rigor, plain-language escalation, project-write boundaries, decision tracking. He liked how that discipline felt and wants Alfred — his always-on assistant across every project — to run with "all of firstmate's best and most vital systems and logic," permanently, not just inside this repo.
+
+I read Alfred's actual global files (`~/.claude/IDENTITY.md`, `SOUL.md`, `USER.md`) to see what's already there before proposing changes. Finding: **a lot of firstmate's discipline is already ported.** Command-center's project `CLAUDE.md` already has "DEFINITION OF DONE — TWO GATES" (Schema-First + Receipt-on-Done, mirroring firstmate's Ground-Truth-Contract), "NO BLIND TURN-END" (mirrors firstmate's "no turn ends blind while work is under way"), full git safety rules, and a structured memory system (mirrors firstmate's `learnings.md`/`captain.md`/`captain-shared.md`). SOUL.md commitment #9 ("Report delegated work in outcomes and decisions, never in watcher/queue/task-id mechanics") is already a condensed version of firstmate's section 9 translation contract.
+
+So this isn't a wholesale import — it's identifying the **genuine gaps**: firstmate discipline that is *not yet* codified anywhere in Alfred's global behavior, cleanly portable (doesn't depend on firstmate's fleet-specific infrastructure — worktrees, herdr/treehouse backends, tasks-axi, crewmate spawn/supervise loop, which have no equivalent surface for Alfred and shouldn't be cargo-culted in without the underlying tech).
+
+## What NOT to port (and why)
+
+- Crew/task lifecycle mechanics (`fm-spawn.sh`, worktree isolation, herdr/treehouse backends, tasks-axi backlog) — this is fleet-orchestration tooling. Alfred's closest equivalent is the `Agent`/`Workflow` tools, which already have their own dispatch/verification conventions (see the Agent tool's "Trust but verify" guidance already in Alfred's system prompt).
+- The full "captain etiquette" jargon-translation table verbatim — firstmate's list translates firstmate-specific terms (worktree, herdr, tasks-axi). Alfred needs the *principle*, not that literal table.
+- Watcher/supervision loop (`fm-watch-arm.sh`) — this exists because firstmate runs a fleet with async background completion the primary session must not miss. Alfred's analogous risk is background `Agent`/`Workflow` dispatch, and command-center's CLAUDE.md **already** has "NO BLIND TURN-END" covering exactly this. No new mechanism needed.
+
+## What to actually add (the real gaps)
+
+1. **Diagnostic-reasoning discipline** — firstmate's rule that a scout/investigation/audit finding is evidence, never authorization to implement; a separate go-ahead is required before code changes. Alfred's global files have nothing like this today — worth a short, universal addition since Alfred routinely produces analyses/audits across every project.
+
+2. **Decision-hold-lifecycle** — firstmate never lets an unresolved decision discovered mid-task quietly vanish; it's recorded and tracked to resolution. Alfred has no equivalent principle stated anywhere. Worth adding as a short universal rule: any open question/decision surfaced mid-task gets explicitly named to Brandon before the task is considered done, not buried in a wall of text or silently dropped.
+
+3. **Sharpen the captain-facing translation contract** — SOUL.md commitment #9 already exists but is narrowly scoped to "delegated work." Broaden it into a general rule: never surface tool/infra mechanics (background task IDs, internal hook names, raw file paths unless needed to act) in conversation with Brandon — always translate to outcome + consequence + next decision. This is the one piece of firstmate's section 9 genuinely worth generalizing.
+
+4. **Knowledge-placement discipline for Alfred's own memory system** — firstmate's `firstmate-coding-guidelines` skill has a real, useful meta-system: a decision tree for where a new fact belongs (inline in the always-loaded file vs. a skill vs. docs vs. script help), plus a "one-owner rule" (state a contract once, cross-reference everywhere else, never duplicate). Alfred's own auto-memory system (described in this session's system prompt) doesn't yet state this discipline explicitly. Worth adding as a short principle so Alfred's memory files stay lean as they accumulate, instead of the CLAUDE.md-bloat problem firstmate had to fix.
+
+## Files to change
+
+- **`~/.claude/SOUL.md`** — extend commitment #9 into the broader translation-contract rule (item 3); add diagnostic-reasoning (item 1) and decision-hold (item 2) as new short commitments (10, 11, 12).
+- **`~/.claude/CLAUDE.md`** — add a short "Knowledge placement discipline" note near the memory-system instructions (item 4), pointing back to SOUL.md for the behavioral commitments so the rule isn't duplicated in two places (practicing the one-owner rule while adding it).
+
+Each addition will be **short** (2-4 lines, matching the existing terse style of SOUL.md's numbered commitments) — not a copy of firstmate's much longer prose, since Alfred's files are deliberately compact and firstmate's own discipline says don't bloat the always-loaded file.
+
+## Verification
+
+- Read back the edited `SOUL.md` and `CLAUDE.md` in full to confirm the new commitments read naturally alongside existing ones, use Alfred's existing voice/style, and don't duplicate content already covered by "DEFINITION OF DONE — TWO GATES" / "NO BLIND TURN-END" in command-center's project CLAUDE.md.
+- No code changes, no repo behavior changes — this is a persona/instruction-file edit only, so "testing" means confirming the files parse as clean markdown and the numbering/cross-references are consistent.

bin/backends/herdr.sh

@@ -2333,8 +2333,8 @@ fm_backend_herdr_events_capable() {  # <session>
   case "$protocol" in ''|*[!0-9]*) return 1 ;; esac
   [ "$protocol" -ge "$FM_BACKEND_HERDR_MIN_EVENTS_PROTOCOL" ] || return 1
   schema=$(herdr api schema --json 2>/dev/null) || return 1
-  printf '%s' "$schema" | grep -Fq 'events.subscribe' || return 1
-  printf '%s' "$schema" | grep -Fq 'pane.agent_status_changed' || return 1
+  case "$schema" in *events.subscribe*) ;; *) return 1 ;; esac
+  case "$schema" in *pane.agent_status_changed*) ;; *) return 1 ;; esac
   return 0
 }
 

bin/fm-brief.sh

@@ -281,6 +281,14 @@ read -r MODE _ <<EOF
 $("$FM_ROOT/bin/fm-project-mode.sh" "$REPO")
 EOF
 
+RUN_EVAL=$(shell_quote "$FM_ROOT/evals/run_eval.sh")
+IFS= read -r -d '' EVAL_STEP <<EOF || true
+Before reporting done, run the relevant eval(s) against your deliverable, e.g. \`$RUN_EVAL completeness <path>\`.
+\`completeness\` applies to every task before Done; add \`tov\` for anything written in Brandon's voice, \`principles\` for a plan/research/recommendation deliverable, \`visual-asset\` for generated imagery, and \`publish-safety\` (last) for anything customer-facing or public.
+A FAIL means loop and fix - never report done anyway.
+EOF
+EVAL_STEP=${EVAL_STEP%$'\n'}
+
 case "$MODE" in
   direct-PR)
     SETUP2=""
@@ -289,6 +297,7 @@ case "$MODE" in
 # Definition of done
 This project ships **direct-PR**: you raise the PR yourself, without the no-mistakes pipeline.
 The task is complete only when committed on your branch.
+$EVAL_STEP
 When it is implemented and committed, push your branch and open a PR with \`gh-axi\`, then append \`done: PR {url}\` to the status file and stop.
 Do NOT run /no-mistakes. The configured merge authority decides whether to merge the PR; firstmate relays the outcome.
 EOF
@@ -301,6 +310,7 @@ EOF
 This project ships **local-only**: no remote, no PR, no pipeline.
 The task is complete only when committed on your branch \`fm/$ID\`. Do NOT push, do NOT open a PR, do NOT merge.
 Keep your branch a clean fast-forward onto the current default branch - if \`main\` has advanced, rebase onto it so the eventual merge stays a fast-forward.
+$EVAL_STEP
 When it is implemented and committed, append \`done: ready in branch fm/$ID\` to the status file and stop.
 The configured merge authority approves the ready branch, then firstmate merges it into local \`main\` through the guarded fast-forward path.
 EOF
@@ -312,6 +322,7 @@ EOF
     IFS= read -r -d '' DOD <<EOF || true
 # Definition of done
 The task is complete only when committed on your branch.
+$EVAL_STEP
 When you believe it is complete, append \`done: {summary}\` to the status file and stop.
 Firstmate will then instruct you to run /no-mistakes to validate and ship a PR.