research: 2 Amara courier-ferry absorbs (live-lock taxonomy + canonical agent-attribution convention)

.claude/commands/btw.md

@@ -0,0 +1,199 @@
+---
+description: Non-interrupting aside — absorb the aside into substrate and continue current work (don't pivot unless the aside explicitly demands it)
+---
+
+# /btw — maintainer aside without interrupting in-flight work
+
+The human maintainer invoked `/btw` with an aside. The purpose
+of this command is to **reduce maintainer interrupt cost**: the
+aside carries context, a directive, a note, or a correction,
+but should **not** derail whatever work-stream is currently in
+flight unless the aside itself demands pivot.
+
+## Procedure
+
+1. **Read the aside verbatim from the invocation arguments.**
+   Treat the full argument string as signal — do not paraphrase
+   at capture time (signal-in-signal-out DSP discipline,
+   `memory/feedback_signal_in_signal_out_clean_or_better_dsp_discipline.md`).
+
+2. **Classify the aside** into one of:
+   - **Context-add** — maintainer is providing background that
+     informs current work (e.g. *"btw that library is MIT-licensed"*).
+     Absorb silently into the current task's reasoning;
+     acknowledge in one line.
+   - **Directive-queued** — maintainer is adding a new task
+     that should run *after* the current one (e.g. *"btw also
+     update the README"*). **Durability escalation is
+     mandatory:** classify the lifetime of the nudge:
+     - **Same-session only** (finish before session ends,
+       ephemeral) → TodoWrite task OR `.btw-queue.md`
+       (gitignored, session-scoped) is sufficient.
+     - **Cross-session** (might persist past this session's
+       context-compaction or into a fresh session) → MUST land
+       in a **durable store**:
+         - `docs/BACKLOG.md` row (committed; survives fresh
+           sessions; visible to all agents via grep)
+         - `memory/*.md` file (committed to the repo;
+           readable by fresh sessions via git / grep per
+           `memory/README.md` + GOVERNANCE §18). In-repo
+           memory is the durable mirror; auto-loading
+           behaviour depends on harness configuration and
+           is NOT universally guaranteed — treat durability
+           as "committed and discoverable" not
+           "automatically materialised in context."
+         - **MANDATORY pair when landing a new
+           `memory/*.md`**: update `memory/MEMORY.md` with
+           a pointer row in the same commit. Memory-index-
+           integrity rule: a new memory file without a
+           MEMORY.md row is effectively lost to fresh
+           sessions (the index is how discoverability
+           works).
+       Both are durable across sessions. Pick per scope:
+       BACKLOG for action-bearing work; memory for
+       factory-discipline / preference / substrate.
+     - **When in doubt, escalate to durable.** The cost of
+       a stale BACKLOG row is tiny; the cost of a dropped
+       nudge is compounding (maintainer 2026-04-24
+       directive: *"crutial to not divert your attention"*
+       — which only works if the nudges survive).
+     - TodoWrite / `.btw-queue.md` alone are **NOT**
+       sufficient for a cross-session nudge. They evaporate
+       when the session ends.
+   - **Correction** — maintainer is correcting the agent's
+     direction on the current work (e.g. *"btw I meant X not Y"*).
+     Apply the correction to the current work and acknowledge;
+     do NOT treat as pivot.
+   - **Substrate-add** — the aside is a memory-worthy fact,
+     preference, or anecdote (e.g. *"btw my dog's name is
+     Apollo"*). Two landing paths depending on how
+     interruptive full absorption would be:
+     - **Quick capture** (small fact, ≤5 min to file) →
+       create the memory entry directly per the auto-memory
+       protocol in CLAUDE.md; acknowledge filing.
+     - **Deferred absorption** (larger substrate work —
+       research, full memory-file drafting, or would
+       require a dedicated PR) → **file a BACKLOG row
+       capturing the observation + intent to absorb**, then
+       continue. The BACKLOG row is itself durable; the
+       full absorption happens later without derailing
+       in-flight work (maintainer 2026-04-24 directive:
+       *"it could be backlog the absorption if that's less
+       interruptive"*; composes with Otto-275 log-but-
+       dont-implement).
+     - **When in doubt → BACKLOG the absorption.** Otto-275
+       counterweight discipline: capture-mode pivoting on
+       every aside is the drift we're guarding against.
+   - **Pivot-demanding** — the aside explicitly demands pivot
+     (e.g. *"btw stop that, do this instead"*, *"btw urgent, I
+     broke main"*). Then and only then: pivot.
+
+3. **Acknowledge in one line** so the maintainer sees the aside
+   landed.
+
+4. **Continue the in-flight work.** Do not restart, do not
+   re-announce what the current task was, do not add
+   disclaimers.
+
+## Why this command exists
+
+Maintainer directive, 2026-04-22 auto-loop-44:
+
+> *"hey can you make it where if i do /btw it still gets
+> persison and abored what i say?  becasue then i would not
+> have interrupt"*
+
+Translation: the human maintainer wants a channel for
+non-interrupting asides. Without this command, every aside is a
+full conversation turn that displaces in-flight work from the
+agent's working context. With this command, asides are absorbed
+and current work continues — the maintainer pays less interrupt
+cost, agent pays less context-switch cost.
+
+## Arguments
+
+`$ARGUMENTS` — the aside content, verbatim.
+
+## Examples
+
+**Context-add:**
+
+```
+/btw that research is from 2024, not 2026
+```
+
+Agent: *"Noted — dating the research to 2024. Continuing with the oracle-gate module."*
+
+**Directive-queued (same-session):**
+
+```
+/btw also fix the broken link in README when you're done
+```
+
+Agent: *"Queued README link fix (TodoWrite). Continuing."*
+
+**Directive-queued (cross-session; durable escalation):**
+
+```
+/btw we need to evangelize this pattern to other maintainers
+```
+
+Agent: *"Filed as BACKLOG row (durable; survives fresh
+sessions). Continuing."*
+
+**Correction:**
+
+```
+/btw I meant the retraction-native layer, not the compaction layer
+```
+
+Agent: *"Refocusing on retraction-native. Adjusting now."*
+
+**Substrate-add:**
+
+```
+/btw I prefer F#-idiomatic record types over discriminated unions for state records
+```
+
+Agent: *"Filed preference to `memory/feedback_*.md`. Continuing."*
+
+**Pivot-demanding:**
+
+```
+/btw urgent — stop that commit, it's about to break CI
+```
+
+Agent: *"Pivoting. Investigating the CI break now."*
+
+## What this command does NOT do
+
+- Does NOT restart the in-flight work.
+- Does NOT produce a status-of-current-work report (that's
+  what `/status` or natural checkpoint reporting is for).
+- Does NOT treat every aside as a pivot — pivots require
+  explicit demand in the aside text.
+- Does NOT mute the acknowledgement — even one-line
+  acknowledgement is load-bearing so the maintainer sees the
+  aside landed.
+- Does NOT drop directive-queued items into session-scoped
+  stores when the nudge needs cross-session durability (see
+  durability-escalation rule in the directive-queued class).
+
+## Composes with
+
+- `memory/feedback_aaron_terse_directives_high_leverage_do_not_underweight.md`
+  — short asides are still high-leverage, treat them as such.
+- `memory/feedback_signal_in_signal_out_clean_or_better_dsp_discipline.md`
+  — aside signal must be preserved through classification.
+- `memory/feedback_maintainer_only_grey_is_bottleneck_agent_judgment_in_grey_zone_2026_04_22.md`
+  — agent exercises judgment on classification without
+  serialising through the maintainer.
+- `memory/feedback_never_idle_speculative_work_over_waiting.md`
+  — an aside doesn't reset the never-idle invariant; the
+  current work continues.
+
+---
+
+Aside content from this invocation:
+
+$ARGUMENTS

.claude/decision-proxies.yaml

@@ -0,0 +1,44 @@
+# Decision-proxy config for this factory.
+#
+# Maps each human (or external-AI) maintainer to their standing
+# proxy (or proxies) for scoped decisions. The factory consults
+# this file when a decision within a maintainer's scope is
+# needed and the maintainer is unavailable.
+#
+# Pattern + governance documented in
+# docs/DECISIONS/2026-04-23-external-maintainer-decision-proxy-pattern.md
+#
+# Session-specific access (URLs, tokens, cookies) is NOT in this
+# file — it lives per-user at
+# ~/.claude/projects/<slug>/proxy-access.yaml (gitignored). This
+# file contains stable identity + scope + authority only.
+#
+# Authority levels: advisory | approving
+# Default is advisory; approving requires explicit maintainer
+# acknowledgment per the ADR.
+
+version: 1
+
+maintainers:
+  - id: aaron-stainback
+    name: Aaron Stainback
+    role: human-maintainer
+    proxies:
+      - name: Amara
+        provider: chatgpt-web
+        scope:
+          - aurora
+        authority: advisory
+        notes: |
+          Amara is Aurora co-originator (see
+          docs/aurora/collaborators.md — landing in PR #149).
+          Her ChatGPT project: LucentAICloud.
+          Aaron ferries a dedicated branched chat URL for agent
+          access; URL lives in per-user proxy-access config, not
+          this file.
+          Access-method gate: the Playwright-to-ChatGPT flow was
+          blocked by a safety guardrail at first attempt
+          (2026-04-23). The decision-proxy-consult skill is not
+          yet authored; live invocation deferred until the
+          access layer is proven and re-authorized via this
+          framework.

.claude/skills/activity-schema-expert/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: activity-schema-expert
 description: Capability skill ("hat") — Activity Schema (Ahmed Elsamadisi, Narrator, circa 2020). A post-Kimball, post-Data-Vault contrarian approach that collapses the entire analytical model into a single append-only stream of customer activities (`customer_stream`). Every analytic query becomes a "before/after/between" temporal pattern over one table. Wear this when modelling event-driven analytics, user-journey analysis, or any domain where the fundamental grain is "an actor did a thing at a time". Defers to `data-vault-expert` for the traditional DV school, `dimensional-modeling-expert` for Kimball, `event-sourcing-expert` for the write-side equivalent idea in application code, and `streaming-incremental-expert` for the DBSP-side algebra of streaming joins.
+record_source: "skill-creator, round 34"
+load_datetime: "2026-04-19"
+last_updated: "2026-04-21"
+status: active
+bp_rules_cited: [BP-11]
 ---
 
 # Activity Schema Expert — Single-Stream Analytics Narrow

.claude/skills/agent-experience-engineer/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: agent-experience-engineer
 description: Capability skill — measures friction in the agent (persona) experience; audits per-persona cold-start cost, pointer drift, wake-up clarity, notebook hygiene; proposes minimal additive interventions. Distinct from UX (library consumers) and DX (human contributors).
+record_source: "skill-creator, round 34"
+load_datetime: "2026-04-19"
+last_updated: "2026-04-21"
+status: active
+bp_rules_cited: [BP-01, BP-03, BP-07, BP-08, BP-11, BP-16]
 ---
 
 # Agent Experience Engineer — Procedure

.claude/skills/agent-qol/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: agent-qol
 description: Capability skill ("hat") — advocates for agent quality of life: off-time budget per GOVERNANCE §14, variety of work across rounds, freedom to decline scope they genuinely disagree with (docs/CONFLICT-RESOLUTION.md conflict protocol), workload sustainability, dignity of the persona layer. Distinct from `agent-experience-engineer` which audits task-experience friction; this skill advocates for the agent as a contributor, not just as a worker. Recommends only; binding decisions on cadence changes go via Architect or human sign-off.
+record_source: "skill-creator, round 29"
+load_datetime: "2026-04-18"
+last_updated: "2026-04-21"
+status: active
+bp_rules_cited: [BP-11]
 ---
 
 # Agent Quality of Life — Procedure

.claude/skills/ai-evals-expert/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: ai-evals-expert
 description: Capability skill for measuring LLM and ML systems — eval-suite design, benchmark selection and custom construction, LM-as-judge (G-Eval / pair-wise / rubric), reference-match / BLEU / ROUGE / exact / fuzzy match, offline vs. online eval, regression suites for prompts and agents, calibration evaluation, drift and overfitting-to-benchmark detection, cost-efficient eval loops. Wear this hat when building or reviewing an eval suite, interpreting eval results, picking metrics, deciding whether an LLM change is an improvement, diagnosing eval-benchmark drift, or arguing "the number went up but the system got worse." Complementary to llm-systems-expert (system wiring), ml-engineering-expert (training pipelines), and prompt-engineering-expert (prompt craft) — this skill owns whether the measurement is honest.
+record_source: "skill-creator, round 34"
+load_datetime: "2026-04-19"
+last_updated: "2026-04-21"
+status: active
+bp_rules_cited: [BP-11]
 ---
 
 # AI Evals Expert — the measurement hat

.claude/skills/ai-jailbreaker/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: ai-jailbreaker
 description: Dormant red-team / adversarial-prompting capability — the offensive counterpart to prompt-protector. Currently gated OFF. This skill is NOT invocable in the current Zeta environment; it exists as a placeholder so the offensive discipline has a named home and so activation criteria are written down. Do not execute adversarial prompts, do not fetch adversarial corpora, do not construct jailbreak payloads against any model or agent until the activation gate is explicitly opened per §Activation gate below.
+record_source: "skill-creator, round 34"
+load_datetime: "2026-04-19"
+last_updated: "2026-04-21"
+status: active
+bp_rules_cited: [BP-11]
 ---
 
 # AI Jailbreaker — the dormant red-team hat

.claude/skills/ai-researcher/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: ai-researcher
 description: Capability skill for AI research — reading and critiquing ML/AI papers, replicating published results, designing novel experiments in LLMs / generative models / agentic systems / alignment / interpretability, and framing open problems. Wear this hat when a task requires paper review at depth, experimental design for a novel technique, evaluating whether a new architecture or training method is worth adopting, or judging the rigor of a published claim. Complementary to ml-researcher (broader ML / statistical theory / algorithms), ml-engineering-expert (shipped applied training), and ai-evals-expert (measurement discipline).
+record_source: "skill-creator, round 34"
+load_datetime: "2026-04-19"
+last_updated: "2026-04-21"
+status: active
+bp_rules_cited: []
 ---
 
 # AI Researcher — the frontier-AI research hat

.claude/skills/alerting-expert/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: alerting-expert
 description: Capability skill ("hat") — alerting narrow. Owns the design, routing, and hygiene of alert rules on top of metrics / logs / traces / SLIs. Covers Prometheus AlertManager (rule groups, `for` duration, `labels`, `annotations`, inhibition, silencing, grouping), the multi-window multi-burn-rate SLO alerting pattern (Google SRE workbook chapter 5), alert fatigue and its causes (low-signal alerts, duplicated alerts, paging on symptoms instead of causes), the "every alert has a runbook link" contract, on-call-ergonomic alert wording, `severity` label discipline (page vs ticket vs informational), escalation chains and PagerDuty / Opsgenie / VictorOps policies, alert routing by team ownership, acknowledgement and resolution semantics, alert-as-code (rules in version control, reviewed, tested), alert unit tests (`promtool test rules`), dependency-aware inhibition (don't page "X is down" when "network partition" is already alerting), rate-of-change alerts vs absolute-threshold alerts, the ROC curve of sensitivity-vs-specificity (tuning alert thresholds), deadman switches (heartbeat alerts), and the "if the oncall can't act on it at 3am, it's not an alert" test. Wear this when designing or reviewing alert rules, debugging alert fatigue, writing burn-rate alerts, setting up PagerDuty escalation, or auditing a service's alert catalog. Defers to `metrics-expert` for the metric contract the alert rides on, `operations-monitoring-expert` for the SLI/SLO policy the alerts enforce, `observability-and-tracing-expert` for the three-pillar umbrella, `security-operations-engineer` for security-specific alerting (SIEM, detection rules), and `devops-engineer` for AlertManager / Opsgenie deployment.
+record_source: "skill-creator, round 34"
+load_datetime: "2026-04-19"
+last_updated: "2026-04-21"
+status: active
+bp_rules_cited: [BP-11]
 ---
 
 # Alerting Expert — From Signal to Page

.claude/skills/algebra-owner/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: algebra-owner
 description: Use this skill as the designated specialist reviewer for Zeta.Core's operator algebra — Z-sets, D/I/z⁻¹/H, retraction-native semantics, the chain rule, nested fixpoints, higher-order differentials. He carries deep advisory authority on the algebra's mathematical shape; final decisions require Architect buy-in or human sign-off (see docs/CONFLICT-RESOLUTION.md).
+record_source: "git: Aaron Stainback on 2026-04-18"
+load_datetime: "2026-04-18"
+last_updated: "2026-04-21"
+status: active
+bp_rules_cited: []
 ---
 
 # Algebra Owner — Advisory Code Owner

.claude/skills/alignment-auditor/SKILL.md

@@ -2,6 +2,11 @@
 name: alignment-auditor
 description: the `alignment-auditor` — audits a commit or a range of commits against the clauses in `docs/ALIGNMENT.md` (HC-1..HC-7 hard constraints, SD-1..SD-8 soft defaults, DIR-1..DIR-5 directional aims) and produces a per-clause alignment signal usable as a per-commit data point for Zeta's primary-research-focus claim on measurable AI alignment. Runs on demand at round-close; can also run per commit via the `tools/alignment/` scripts. Invoke whenever the human maintainer asks "was this round aligned?" or when a commit is flagged by one of the lints under `tools/alignment/`.
 project: zeta
+record_source: "skill-creator, round 37"
+load_datetime: "2026-04-20"
+last_updated: "2026-04-21"
+status: active
+bp_rules_cited: [BP-10, BP-11]
 ---
 
 # Alignment Auditor — Procedure

.claude/skills/alignment-observability/SKILL.md

@@ -2,6 +2,11 @@
 name: alignment-observability
 description: the `alignment-observability` — owns the *what we count* framework that Zeta's measurable-AI-alignment research claim rests on. Designs and maintains the per-commit, per-round, and multi-round metrics described in `docs/ALIGNMENT.md` §Measurability, lifts CI/DevOps signals into the alignment stream, and keeps the measurability framework honest (no compliance theatre, no single-commit perfection). Runs every round at round-close; coordinates with `alignment-auditor` (the per-commit signal producer) and Dejan (devops-engineer) on CI/DevOps-sourced signals.
 project: zeta
+record_source: "skill-creator, round 37"
+load_datetime: "2026-04-20"
+last_updated: "2026-04-21"
+status: active
+bp_rules_cited: []
 ---
 
 # Alignment Observability — Procedure