aether-colony 5.3.2 → 5.4.0

package/.aether/docs/source-of-truth-map.md

@@ -36,10 +36,10 @@ Define which files are authoritative for system behavior, which files are derive
 |---|---|---|
 | Core deterministic operations | `.aether/aether-utils.sh` | Dispatcher that loads domain modules on demand (~5,200 lines) |
 | Slash command orchestration | `.claude/commands/ant/*.md` | Includes build/continue orchestrators |
-| Worker behavior specs | `.claude/agents/ant/*.md` | 22 Claude agent definitions (Builder, Watcher, etc.) |
+| Worker behavior specs | `.claude/agents/ant/*.md` | 24 Claude agent definitions (Builder, Watcher, etc.) |
 | Packaged Claude agent mirror | `.aether/agents-claude/*.md` | Distribution mirror; must stay byte-identical with `.claude/agents/ant/*.md` |
-| OpenCode command surface | `.opencode/commands/ant/*.md` | 44 OpenCode command files; structure parity with Claude commands |
-| OpenCode worker behavior specs | `.opencode/agents/*.md` | 22 OpenCode agent definitions |
+| OpenCode command surface | `.opencode/commands/ant/*.md` | 45 OpenCode command files; structure parity with Claude commands |
+| OpenCode worker behavior specs | `.opencode/agents/*.md` | 24 OpenCode agent definitions |
 | Domain modules | `.aether/utils/{flag,spawn,session,suggest,queen,swarm,learning,pheromone,state-api}.sh` | Extracted from aether-utils.sh in Phase 13; sourced on demand |
 | Build/continue split stages | `.aether/docs/command-playbooks/*.md` | Loaded by orchestrators; executable instruction docs |
 | Output templates | `.aether/templates/*` | Templates for generated state/handoff/wisdom/session artifacts |
@@ -66,8 +66,8 @@ Define which files are authoritative for system behavior, which files are derive
 - `build.md` and `continue.md` are now orchestrators that load split playbooks under `.aether/docs/command-playbooks/`.
 - Orchestrators run playbooks as staged instruction sets (Read-tool execution model), not as bash subcommand wrappers.
 - Cross-platform surfaces are present:
-  - 44 Claude commands and 44 OpenCode commands
-  - 22 Claude agents and 22 OpenCode agents
+  - 45 Claude commands and 45 OpenCode commands
+  - 24 Claude agents and 24 OpenCode agents
 - Autopilot subcommands (`autopilot-init`, `autopilot-update`, `autopilot-status`, `autopilot-stop`, `autopilot-check-replan`) manage autonomous build/continue cycles via `/ant:run`.
 - `registry-list` lists all registered repos with metadata including domain tags.
 - `.aether/agents-claude/*.md` mirrors `.claude/agents/ant/*.md` for packaging/distribution.
@@ -84,11 +84,11 @@ Define which files are authoritative for system behavior, which files are derive
 | Core utility entrypoint | `.aether/aether-utils.sh` | 1 | Active |
 | Sourced shell utilities | `.aether/utils/*.sh` | ~29 | Active (9 domain modules + infrastructure + XML) |
 | XML utility scripts | `.aether/utils/xml-*.sh` | 5 | Active (see drift note) |
-| Slash commands (Claude) | `.claude/commands/ant/*.md` | 44 | Active |
-| Slash commands (OpenCode) | `.opencode/commands/ant/*.md` | 44 | Active (content differs from Claude variants) |
-| Agent definitions (Claude) | `.claude/agents/ant/*.md` | 22 | Active |
-| Agent mirror (packaging) | `.aether/agents-claude/*.md` | 22 | Active mirror (must match Claude agent files exactly) |
-| Agent definitions (OpenCode) | `.opencode/agents/*.md` | 22 | Active (content differs from Claude variants) |
+| Slash commands (Claude) | `.claude/commands/ant/*.md` | 45 | Active |
+| Slash commands (OpenCode) | `.opencode/commands/ant/*.md` | 45 | Active (content differs from Claude variants) |
+| Agent definitions (Claude) | `.claude/agents/ant/*.md` | 24 | Active |
+| Agent mirror (packaging) | `.aether/agents-claude/*.md` | 24 | Active mirror (must match Claude agent files exactly) |
+| Agent definitions (OpenCode) | `.opencode/agents/*.md` | 24 | Active (content differs from Claude variants) |
 | Command playbooks | `.aether/docs/command-playbooks/*.md` | 9 | Active (5 build + 4 continue; excludes full/README) |
 | Templates (all types) | `.aether/templates/*` | 12 | Active |
 | Disciplines | `.aether/docs/disciplines/*.md` | 7 | Active |

package/.aether/docs/structural-learning-stack.md

@@ -0,0 +1,283 @@
+# Structural Learning Stack
+
+> Memory consolidation pipeline: from raw observations to trusted wisdom.
+
+---
+
+## Overview
+
+The Structural Learning Stack is how Aether converts raw, repeated observations into
+colony wisdom that workers can act on. Each stage adds structure: trust scores quantify
+reliability, the graph records relationships between instincts, curation ants clean and
+promote, and the event bus connects everything loosely.
+
+The stack runs automatically at phase-end and seal — workers never call it directly.
+
+---
+
+## Architecture Diagram
+
+```
+                          learning-observe
+                               │
+                    [learning-observations.json]
+                               │
+                         trust-calculate
+                               │
+                               ▼
+                    [instinct-store] ──────────► [instinct-graph.json]
+                               │                  (graph-link edges)
+                               │
+                    ┌──────────▼──────────┐
+                    │   Curation Ants     │
+                    │  (phase-end / seal) │
+                    └──────────┬──────────┘
+                               │
+               ┌───────────────┼───────────────┐
+               ▼               ▼               ▼
+          [QUEEN.md]    [event-bus.jsonl]  [archived]
+        (via herald)    (consolidation     (low-trust
+                         events)           instincts)
+```
+
+---
+
+## Pipeline Stages
+
+### 1. Observation (learning.sh — `_learning_observe`)
+
+Records an observation of a learning pattern into `.aether/data/learning-observations.json`.
+
+- **Deduplication:** A SHA-256 hash of the content is used as the key. If an identical observation already exists, `observation_count` is incremented and `last_seen` is updated rather than creating a duplicate entry.
+- **Trust score on write:** For new observations, `trust-calculate` is called immediately using the provided `source_type` and `evidence_type` (defaults: `observation` / `anecdotal`). The computed score is stored as `trust_score`.
+- **Backup rotation:** A 3-file rotating backup (`*.bak.1`, `*.bak.2`, `*.bak.3`) is maintained on every write. On read, if the file is corrupt, the most recent valid backup is restored automatically.
+- **Valid wisdom types:** `philosophy`, `pattern`, `redirect`, `stack`, `decree`, `failure`
+
+Observation schema fields: `content_hash`, `content`, `wisdom_type`, `observation_count`, `first_seen`, `last_seen`, `colonies`, `trust_score`, `source_type`, `evidence_type`, `compression_level`.
+
+---
+
+### 2. Trust Scoring Engine (trust-scoring.sh)
+
+A pure calculation module. No state is read or written. All functions accept `--flag value` arguments and return JSON.
+
+**Weighted formula:**
+```
+score = 0.40 × source_score + 0.35 × evidence_score + 0.25 × activity_score
+```
+
+**Source types (40% weight):**
+
+| Type | Score |
+|------|-------|
+| user_feedback | 1.0 |
+| error_resolution | 0.9 |
+| success_pattern | 0.8 |
+| observation | 0.6 |
+| heuristic | 0.4 |
+
+**Evidence types (35% weight):**
+
+| Type | Score |
+|------|-------|
+| test_verified | 1.0 |
+| multi_phase | 0.9 |
+| single_phase | 0.7 |
+| anecdotal | 0.4 |
+
+**Activity score (25% weight):** `0.5 ^ (days_since_last_use / 60)` — a 60-day half-life decay from the last observed use.
+
+**Floor:** Score is never below `0.2`.
+
+**Trust tiers:**
+
+| Score range | Tier | Index |
+|-------------|------|-------|
+| 0.90 – 1.00 | canonical | 0 |
+| 0.80 – 0.89 | trusted | 1 |
+| 0.70 – 0.79 | established | 2 |
+| 0.60 – 0.69 | emerging | 3 |
+| 0.45 – 0.59 | provisional | 4 |
+| 0.30 – 0.44 | suspect | 5 |
+| 0.20 – 0.29 | dormant | 6 |
+
+---
+
+### 3. Event Bus (event-bus.sh)
+
+A JSONL append-log at `.aether/data/event-bus.jsonl`. Each line is one JSON event object.
+
+**Event fields:** `id`, `topic`, `payload`, `source`, `timestamp`, `ttl_days`, `expires_at`
+
+- **Pub/sub pattern:** Publish to a topic with `event-publish`; subscribe by topic pattern with `event-subscribe`. Patterns support exact match or prefix wildcard (e.g., `"consolidation.*"`).
+- **TTL:** Default 30 days per event. Events with `expires_at` in the past are excluded from subscriptions and removed by cleanup.
+- **Concurrency:** File locking via `acquire_lock`/`release_lock` on every append and cleanup rewrite.
+- **Replay:** `event-replay` returns events for a topic from a given ISO-8601 timestamp, sorted chronologically.
+- **Cleanup:** `event-cleanup` atomically rewrites the JSONL file, keeping only non-expired events. Supports `--dry-run`.
+
+---
+
+### 4. Instinct Storage (instinct-store.sh)
+
+Persistent trust-scored instincts at `.aether/data/instincts.json` (schema version 1.0).
+
+**Instinct schema:**
+```json
+{
+  "id": "inst_<timestamp>_<hex>",
+  "trigger": "...",
+  "action": "...",
+  "domain": "...",
+  "trust_score": 0.75,
+  "trust_tier": "established",
+  "confidence": 0.8,
+  "provenance": {
+    "source": "...",
+    "source_type": "observation",
+    "evidence": "...",
+    "created_at": "<ISO8601>",
+    "last_applied": null,
+    "application_count": 0
+  },
+  "application_history": [],
+  "related_instincts": [],
+  "archived": false
+}
+```
+
+- **Deduplication:** First 50 characters of `trigger` are matched against existing active entries. On match, confidence is boosted to `max(existing, new)` and the trust score is recomputed.
+- **50-instinct cap:** When active count exceeds 50 after insert, the entry with the lowest `trust_score` is soft-deleted (`archived: true`).
+- **Decay:** `instinct-decay-all` applies the same 60-day half-life decay to all active entries. Entries whose decayed score falls below `0.25` are automatically archived.
+- **Read:** `instinct-read-trusted` returns active entries filtered by `min-score` (default 0.5), optionally by domain, sorted by `trust_score` descending.
+
+---
+
+### 5. Graph Layer (graph.sh)
+
+Tracks directed relationships between instincts at `.aether/data/instinct-graph.json`.
+
+**Edge schema:** `edge_id`, `source`, `target`, `relationship`, `weight`, `created_at`
+
+**Relationship types:** `reinforces`, `contradicts`, `extends`, `supersedes`, `related`
+
+**Operations:**
+
+| Function | Description |
+|----------|-------------|
+| `graph-link` | Create or update a directed edge. Duplicate `source+target+relationship` updates the weight. Default weight: 0.5. |
+| `graph-neighbors` | 1-hop lookup. Direction: `out`, `in`, or `both`. Optional relationship filter. |
+| `graph-reach` | BFS traversal up to N hops (max 3). Returns `{id, hop, path}` for each reachable node. Optional `--min-weight` filter. Early exit when no new nodes are found. |
+| `graph-cluster` | Finds clusters of strongly connected instincts using greedy union-find over qualifying edges. Defaults: `min-edges=2`, `min-weight=0.3`. Returns `{nodes, edge_count, avg_weight}` per cluster. |
+
+---
+
+### 6. Curation Ants
+
+Eight specialized ants that maintain memory quality. The orchestrator (`curation-run`) runs all eight in sequence. Sentinel can abort the run early if critical corruption is detected.
+
+**Execution order:**
+
+```
+sentinel → nurse → critic → herald → janitor → archivist → librarian → scribe
+```
+
+**Ant responsibilities:**
+
+| Ant | Responsibility |
+|-----|---------------|
+| sentinel | Health check on all memory stores. Aborts subsequent steps if critical corruption is found. |
+| nurse | Recalculates trust scores for observations and instincts using current elapsed days. |
+| critic | Detects contradictions between instincts (overlapping triggers with conflicting actions). |
+| herald | Promotes high-trust instincts to QUEEN.md Patterns/Philosophies section. |
+| janitor | Removes expired events from the event bus; prunes old archived instincts. |
+| archivist | Archives active instincts whose trust score has fallen below a configurable threshold (default: 0.3 at seal). |
+| librarian | Collects inventory statistics across all memory stores. |
+| scribe | Generates a markdown consolidation report. |
+
+All steps are non-blocking — a failure in one step is logged and execution continues.
+
+---
+
+### 7. Consolidation Pipeline
+
+Two consolidation modes, each calling into the curation ants.
+
+**Phase-end (lightweight) — `consolidation-phase-end`:**
+
+Runs at the end of every phase (`/ant:continue`). Executes three ants only: `nurse → herald → janitor`. Publishes a `consolidation.phase_end` event on the event bus. All three steps are non-blocking.
+
+**Seal (full) — `consolidation-seal`:**
+
+Runs once during `/ant:seal`. Executes five steps:
+
+1. `curation-run` — full 8-ant orchestration
+2. `instinct-decay-all` — final trust decay pass across all active instincts
+3. `curation-archivist --threshold 0.3` — archive borderline instincts
+4. `event-publish` — publish `consolidation.seal` event
+5. `curation-scribe` — generate final consolidation report
+
+All steps are non-blocking. The seal report path is returned in the output.
+
+---
+
+## Integration Points
+
+| Trigger | Stack call | Effect |
+|---------|-----------|--------|
+| `/ant:continue` | `consolidation-phase-end` | nurse + herald + janitor; phase_end event |
+| `/ant:seal` | `consolidation-seal` | full 8-ant curation + decay + archive + seal event + report |
+| `/ant:build` (pattern capture) | `learning-observe` | Records observation with trust score |
+| `colony-prime` | `instinct-read-trusted` | Injects trusted instincts into worker prompts |
+
+---
+
+## Subcommand Reference
+
+| Subcommand | Module | Usage |
+|-----------|--------|-------|
+| `trust-calculate` | trust-scoring.sh | `--source <type> --evidence <type> --days-since <N>` |
+| `trust-decay` | trust-scoring.sh | `--score <float> --days <N>` |
+| `trust-tier` | trust-scoring.sh | `--score <float>` |
+| `event-publish` | event-bus.sh | `--topic <topic> --payload <json> [--source <src>] [--ttl <days>]` |
+| `event-subscribe` | event-bus.sh | `--topic <pattern> [--since <ISO>] [--limit <N>]` |
+| `event-cleanup` | event-bus.sh | `[--dry-run]` |
+| `event-replay` | event-bus.sh | `--topic <topic> --since <ISO> [--limit <N>]` |
+| `instinct-store` | instinct-store.sh | `--trigger <t> --action <a> --domain <d> --confidence <f> --source <s> --evidence <e> [--source-type <type>]` |
+| `instinct-read-trusted` | instinct-store.sh | `[--min-score <f>] [--domain <d>] [--limit <N>]` |
+| `instinct-decay-all` | instinct-store.sh | `[--days <N>] [--dry-run]` |
+| `instinct-archive` | instinct-store.sh | `--id <id>` |
+| `graph-link` | graph.sh | `--source <id> --target <id> --relationship <type> [--weight <float>]` |
+| `graph-neighbors` | graph.sh | `--id <id> [--direction out\|in\|both] [--relationship <type>]` |
+| `graph-reach` | graph.sh | `--id <id> --hops <N> [--min-weight <float>]` |
+| `graph-cluster` | graph.sh | `[--min-edges <N>] [--min-weight <float>]` |
+| `curation-sentinel` | curation-ants/sentinel.sh | `[--dry-run]` |
+| `curation-nurse` | curation-ants/nurse.sh | `[--dry-run]` |
+| `curation-critic` | curation-ants/critic.sh | `[--dry-run]` |
+| `curation-herald` | curation-ants/herald.sh | `[--dry-run]` |
+| `curation-janitor` | curation-ants/janitor.sh | `[--dry-run]` |
+| `curation-archivist` | curation-ants/archivist.sh | `[--threshold <f>] [--dry-run]` |
+| `curation-librarian` | curation-ants/librarian.sh | `[--dry-run]` |
+| `curation-scribe` | curation-ants/scribe.sh | `[--dry-run]` |
+| `curation-run` | curation-ants/orchestrator.sh | `[--dry-run] [--verbose]` |
+| `consolidation-phase-end` | consolidation.sh | `[--dry-run]` |
+| `consolidation-seal` | consolidation-seal.sh | `[--dry-run]` |
+
+---
+
+## Test Coverage
+
+| Test file | What it covers |
+|-----------|---------------|
+| `tests/bash/test-trust-scoring.sh` | `trust-calculate`, `trust-decay`, `trust-tier`; formula weights, floor, tiers |
+| `tests/bash/test-event-bus.sh` | `event-publish`, `event-subscribe`, `event-cleanup`, `event-replay`; TTL, locking, wildcard patterns |
+| `tests/bash/test-instinct-store.sh` | `instinct-store`, `instinct-read-trusted`, `instinct-decay-all`, `instinct-archive`; 50-cap, dedup, decay archival |
+| `tests/bash/test-instinct-apply.sh` | Instinct application and provenance tracking |
+| `tests/bash/test-graph.sh` | `graph-link`, `graph-neighbors`, `graph-reach`, `graph-cluster`; BFS traversal, cluster detection |
+| `tests/bash/test-curation-core.sh` | Core curation ant behavior (sentinel, nurse, critic) |
+| `tests/bash/test-curation-ops.sh` | Operational curation ants (herald, janitor, archivist, librarian, scribe) |
+| `tests/bash/test-curation-orchestrator.sh` | `curation-run` full sequence, sentinel abort path |
+| `tests/bash/test-consolidation.sh` | `consolidation-phase-end`; phase-end step sequence, event publishing |
+| `tests/bash/test-consolidation-seal.sh` | `consolidation-seal`; full seal sequence, dry-run mode |
+| `tests/bash/test-learning-module.sh` | `learning-observe`; trust score on write, deduplication, observation_count increment |
+| `tests/bash/test-learning-recovery.sh` | Backup rotation and corrupt-file recovery logic |
+| `tests/bash/test-oracle-trust.sh` | Trust scoring integration with oracle/research paths |

package/.aether/templates/colony-state-template.json

@@ -0,0 +1 @@
+{}

package/.aether/utils/consolidation-seal.sh

@@ -0,0 +1,196 @@
+#!/bin/bash
+# Consolidation Seal — Full seal-ceremony consolidation
+# Provides: _consolidation_seal
+#
+# Runs once during /ant:seal. Orchestrates curation-run, instinct-decay-all,
+# curation-archivist, event publish, and curation-scribe in sequence.
+# Each step is non-blocking: failures are logged but do not stop the seal.
+#
+# These functions are sourced by aether-utils.sh at startup.
+# All shared infrastructure (json_ok, json_err, COLONY_DATA_DIR, DATA_DIR,
+# error constants) is available when sourced.
+
+# ============================================================================
+# _consolidation_seal
+# Full consolidation for the seal ceremony.
+#
+# Usage: consolidation-seal [--dry-run]
+#
+# Steps:
+#   1. curation-run          — full 8-ant orchestration
+#   2. instinct-decay-all    — final trust decay pass
+#   3. curation-archivist    — archive borderline instincts (threshold 0.3)
+#   4. event-publish         — publish consolidation.seal event
+#   5. curation-scribe       — generate final report
+#
+# Output: json_ok with {type, steps, event_published, dry_run}
+# ============================================================================
+_consolidation_seal() {
+    local dry_run="false"
+
+    while [[ $# -gt 0 ]]; do
+        case "$1" in
+            --dry-run) dry_run="true"; shift ;;
+            *)         shift ;;
+        esac
+    done
+
+    local steps_json="[]"
+    local event_published="false"
+
+    # _cs_step <name> <subcommand> [extra_args...]
+    # Runs the subcommand non-blocking (failures logged, not fatal).
+    # Appends a step entry to steps_json.
+    _cs_step() {
+        local step_name="$1"
+        shift
+        local subcmd="$1"
+        shift
+        # Remaining positional args are passed to the subcommand
+
+        local step_result step_status step_summary step_report_path="null"
+
+        # Build argument array respecting dry_run
+        local -a args=("$subcmd")
+        if [[ "$dry_run" == "true" ]]; then
+            args+=("--dry-run")
+        fi
+        # Append any extra args
+        while [[ $# -gt 0 ]]; do
+            args+=("$1")
+            shift
+        done
+
+        step_result=$(bash "$0" "${args[@]}" 2>/dev/null) || true
+
+        if echo "$step_result" | jq -e '.ok == true' >/dev/null 2>&1; then
+            step_status="ok"
+            step_summary=$(echo "$step_result" | jq -r '
+                .result |
+                if type == "object" then
+                    [ to_entries[] | "\(.key) \(.value)" ] | join(", ")
+                else
+                    tostring
+                end
+            ' 2>/dev/null | head -c 120 || echo "ok")
+            [[ -z "$step_summary" ]] && step_summary="ok"
+
+            # Extract report_path if present (scribe step)
+            step_report_path=$(echo "$step_result" | jq -r '.result.report_path // "null"' 2>/dev/null || echo "null")
+        else
+            step_status="failed"
+            local err_msg
+            err_msg=$(echo "$step_result" | jq -r '.error // "unknown error"' 2>/dev/null || echo "unknown error")
+            step_summary="$err_msg"
+        fi
+
+        steps_json=$(echo "$steps_json" | jq \
+            --arg name "$step_name" \
+            --arg status "$step_status" \
+            --arg summary "$step_summary" \
+            --arg rp "$step_report_path" \
+            '. += [{
+                "name":        $name,
+                "status":      $status,
+                "summary":     $summary,
+                "report_path": (if $rp == "null" then null else $rp end)
+            }]')
+    }
+
+    # ── Step 1: Full 8-ant curation run ────────────────────────────────────────
+    _cs_step "curation-run" "curation-run"
+
+    # ── Step 2: Final trust decay pass ─────────────────────────────────────────
+    _cs_step "instinct-decay-all" "instinct-decay-all"
+
+    # ── Step 3: Archive borderline instincts (threshold 0.3) ───────────────────
+    # Note: --dry-run is injected by _cs_step; but archivist also takes --threshold
+    # We cannot let _cs_step inject --dry-run before --threshold so we handle manually
+    local arch_result arch_status arch_summary
+    local -a arch_args=("curation-archivist" "--threshold" "0.3")
+    if [[ "$dry_run" == "true" ]]; then
+        arch_args+=("--dry-run")
+    fi
+    arch_result=$(bash "$0" "${arch_args[@]}" 2>/dev/null) || true
+    if echo "$arch_result" | jq -e '.ok == true' >/dev/null 2>&1; then
+        arch_status="ok"
+        arch_summary=$(echo "$arch_result" | jq -r '
+            .result |
+            if type == "object" then
+                [ to_entries[] | "\(.key) \(.value)" ] | join(", ")
+            else tostring end
+        ' 2>/dev/null | head -c 120 || echo "ok")
+        [[ -z "$arch_summary" ]] && arch_summary="ok"
+    else
+        arch_status="failed"
+        arch_summary=$(echo "$arch_result" | jq -r '.error // "unknown error"' 2>/dev/null || echo "unknown error")
+    fi
+    steps_json=$(echo "$steps_json" | jq \
+        --arg status "$arch_status" \
+        --arg summary "$arch_summary" \
+        '. += [{"name":"archivist","status":$status,"summary":$summary,"report_path":null}]')
+
+    # ── Step 4: Publish consolidation.seal event ───────────────────────────────
+    if [[ "$dry_run" != "true" ]]; then
+        local payload
+        payload=$(jq -nc \
+            --arg ts "$(date -u +"%Y-%m-%dT%H:%M:%SZ")" \
+            --argjson steps "$steps_json" \
+            '{"timestamp":$ts,"steps":$steps}')
+        local ep_result
+        ep_result=$(bash "$0" event-publish \
+            --topic "consolidation.seal" \
+            --payload "$payload" \
+            --source "consolidation-seal" 2>/dev/null) || true
+        if echo "$ep_result" | jq -e '.ok == true' >/dev/null 2>&1; then
+            event_published="true"
+        fi
+    else
+        # In dry-run mode, report event as published (no side effects)
+        event_published="true"
+    fi
+
+    # ── Step 5: Generate final scribe report ───────────────────────────────────
+    local scribe_result scribe_status scribe_summary scribe_report_path="null"
+    local -a scribe_args=("curation-scribe")
+    if [[ "$dry_run" == "true" ]]; then
+        scribe_args+=("--dry-run")
+    fi
+    scribe_result=$(bash "$0" "${scribe_args[@]}" 2>/dev/null) || true
+    if echo "$scribe_result" | jq -e '.ok == true' >/dev/null 2>&1; then
+        scribe_status="ok"
+        scribe_summary=$(echo "$scribe_result" | jq -r '
+            .result |
+            if type == "object" then
+                [ to_entries[] | "\(.key) \(.value)" ] | join(", ")
+            else tostring end
+        ' 2>/dev/null | head -c 120 || echo "ok")
+        [[ -z "$scribe_summary" ]] && scribe_summary="ok"
+        scribe_report_path=$(echo "$scribe_result" | jq -r '.result.report_path // "null"' 2>/dev/null || echo "null")
+    else
+        scribe_status="failed"
+        scribe_summary=$(echo "$scribe_result" | jq -r '.error // "unknown error"' 2>/dev/null || echo "unknown error")
+    fi
+    steps_json=$(echo "$steps_json" | jq \
+        --arg status "$scribe_status" \
+        --arg summary "$scribe_summary" \
+        --arg rp "$scribe_report_path" \
+        '. += [{
+            "name":        "scribe",
+            "status":      $status,
+            "summary":     $summary,
+            "report_path": (if $rp == "null" then null else $rp end)
+        }]')
+
+    # ── Output ─────────────────────────────────────────────────────────────────
+    json_ok "$(jq -n \
+        --argjson steps "$steps_json" \
+        --argjson event_published "$([ "$event_published" == "true" ] && echo true || echo false)" \
+        --argjson dry "$([ "$dry_run" == "true" ] && echo true || echo false)" \
+        '{
+            "type":            "seal",
+            "steps":           $steps,
+            "event_published": $event_published,
+            "dry_run":         $dry
+        }')"
+}

package/.aether/utils/consolidation.sh

@@ -0,0 +1,127 @@
+#!/bin/bash
+# Consolidation utility — lightweight end-of-phase consolidation
+# Provides: _consolidation_phase_end
+#
+# These functions are sourced by aether-utils.sh at startup.
+# All shared infrastructure (json_ok, json_err, atomic_write,
+# COLONY_DATA_DIR, SCRIPT_DIR, error constants) is available.
+#
+# Subcommand: consolidation-phase-end [--dry-run]
+# Runs nurse + herald + janitor curation ants in sequence,
+# then publishes a consolidation.phase_end event to the event bus.
+# Each step is non-blocking: failures are logged and execution continues.
+
+# ============================================================================
+# _consolidation_phase_end
+# Run lightweight end-of-phase consolidation.
+#
+# Usage: consolidation-phase-end [--dry-run]
+#
+# Output: json_ok with:
+#   {type, steps:[{name, status, summary}], event_published, dry_run}
+# ============================================================================
+_consolidation_phase_end() {
+    local dry_run="false"
+
+    while [[ $# -gt 0 ]]; do
+        case "$1" in
+            --dry-run) dry_run="true"; shift ;;
+            *) shift ;;
+        esac
+    done
+
+    local dry_run_flag=""
+    [[ "$dry_run" == "true" ]] && dry_run_flag="--dry-run"
+
+    local steps_json="[]"
+
+    # ── Step 1: curation-nurse (trust recalculation) ──────────────────────────
+    local nurse_status="ok"
+    local nurse_summary=""
+    local nurse_result
+    # shellcheck disable=SC2086
+    nurse_result=$(bash "$0" curation-nurse $dry_run_flag 2>/dev/null) || nurse_status="failed"
+
+    if [[ "$nurse_status" == "ok" ]]; then
+        nurse_summary=$(echo "$nurse_result" | jq -r \
+            '"observations_updated=\(.result.observations_updated // 0) instincts_updated=\(.result.instincts_updated // 0)"' \
+            2>/dev/null || echo "completed")
+    else
+        nurse_summary="nurse step failed; continuing"
+    fi
+
+    steps_json=$(echo "$steps_json" | jq -c \
+        --arg name "nurse" \
+        --arg status "$nurse_status" \
+        --arg summary "$nurse_summary" \
+        '. + [{name:$name,status:$status,summary:$summary}]')
+
+    # ── Step 2: curation-herald (promote high-trust to QUEEN.md) ─────────────
+    local herald_status="ok"
+    local herald_summary=""
+    local herald_result
+    # shellcheck disable=SC2086
+    herald_result=$(bash "$0" curation-herald $dry_run_flag 2>/dev/null) || herald_status="failed"
+
+    if [[ "$herald_status" == "ok" ]]; then
+        herald_summary=$(echo "$herald_result" | jq -r \
+            '"eligible=\(.result.eligible // 0) promoted=\(.result.promoted // 0)"' \
+            2>/dev/null || echo "completed")
+    else
+        herald_summary="herald step failed; continuing"
+    fi
+
+    steps_json=$(echo "$steps_json" | jq -c \
+        --arg name "herald" \
+        --arg status "$herald_status" \
+        --arg summary "$herald_summary" \
+        '. + [{name:$name,status:$status,summary:$summary}]')
+
+    # ── Step 3: curation-janitor (clean expired) ─────────────────────────────
+    local janitor_status="ok"
+    local janitor_summary=""
+    local janitor_result
+    # shellcheck disable=SC2086
+    janitor_result=$(bash "$0" curation-janitor $dry_run_flag 2>/dev/null) || janitor_status="failed"
+
+    if [[ "$janitor_status" == "ok" ]]; then
+        janitor_summary=$(echo "$janitor_result" | jq -r \
+            '"events_removed=\(.result.events_removed // 0) instincts_pruned=\(.result.instincts_pruned // 0)"' \
+            2>/dev/null || echo "completed")
+    else
+        janitor_summary="janitor step failed; continuing"
+    fi
+
+    steps_json=$(echo "$steps_json" | jq -c \
+        --arg name "janitor" \
+        --arg status "$janitor_status" \
+        --arg summary "$janitor_summary" \
+        '. + [{name:$name,status:$status,summary:$summary}]')
+
+    # ── Step 4: publish event to event bus ───────────────────────────────────
+    local event_published="false"
+
+    local current_phase
+    current_phase=$(jq -r '.current_phase // 0' "$COLONY_DATA_DIR/COLONY_STATE.json" 2>/dev/null || echo "0")
+
+    local payload
+    payload=$(jq -nc \
+        --argjson phase "$current_phase" \
+        --argjson dry_run "$dry_run" \
+        --argjson steps "$steps_json" \
+        '{phase:$phase,dry_run:$dry_run,steps:$steps}')
+
+    if bash "$0" event-publish \
+            --topic "consolidation.phase_end" \
+            --payload "$payload" \
+            --source "consolidation" \
+            > /dev/null 2>&1; then
+        event_published="true"
+    fi
+
+    json_ok "$(jq -nc \
+        --argjson steps "$steps_json" \
+        --argjson event_published "$event_published" \
+        --argjson dry_run "$dry_run" \
+        '{type:"phase_end",steps:$steps,event_published:$event_published,dry_run:$dry_run}')"
+}