chore(skills): propagate lifecycle dependency closure from 321e721

.agents/skills/README.md

@@ -3,51 +3,84 @@
 This directory contains repo-local copies of canonical skills from
 `The-Interdependency/skill-lib`.
 
-Source commit: `38c64332b840b2bbe1c07e53aeee8996644548e9`
+Source commit: `321e721fb660e0fea5bb8c37d943611963039a57`
 
 Repo-local copies are not the source of truth. Edit `skill-lib` first,
 then propagate from the canonical source.
 
 Skills refreshed from the source commit above:
 
+- `wdll/`
 - `a0p-instancing/`
-- `action-calibration/`
 - `agent-instantiation/`
-- `canon/`
-- `cap-build/`
-- `char-compress/`
-- `data-visualization/`
-- `deps-build/`
-- `distributed-publication/`
-- `doc-build/`
-- `domain-claims/`
-- `epac-selection-display/`
-- `explore-data/`
-- `fresh-making/`
-- `gonol-build/`
-- `interdependent-work-graph/`
-- `llms-build/`
-- `loop-eng/`
-- `manifest/`
-- `meta/`
-- `meta-module-build/`
-- `msdmd/`
-- `owner-build/`
-- `plain-lens/`
-- `project-incubation-graduation/`
-- `ratios/`
-- `repo-audit-repair/`
-- `risk-boundary-build/`
-- `skill-build/`
-- `skill-usage/`
-- `sql-queries/`
-- `ssh-automation/`
-- `statistical-analysis/`
-- `test-build/`
-- `the-interdependency/`
-- `thought-lens/`
-- `typed-meta-frontend/`
-- `ucns-option-selection/`
-- `validate-data/`
-- `visitor-intro/`
-- `vm-mcp/`
+
+Other installed skills (not refreshed by this propagation):
+
+Their existing files retain their own authority/provenance; the source
+commit above does not assert that these copies were refreshed.
+
+- `action-calibration/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `canon/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `cap-build/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `char-compress/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `data-visualization/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `deps-build/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `distributed-publication/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `doc-build/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `domain-claims/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `epac-selection-display/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `explore-data/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `fresh-making/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `gonol-build/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `interdependent-work-graph/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `llms-build/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `loop-eng/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `manifest/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `meta/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `meta-module-build/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `msdmd/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `owner-build/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `plain-lens/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `project-incubation-graduation/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `ratios/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `repo-audit-repair/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `risk-boundary-build/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `skill-build/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `skill-usage/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `sql-queries/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `ssh-automation/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `statistical-analysis/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `test-build/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `the-interdependency/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `thought-lens/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `typed-meta-frontend/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `ucns-option-selection/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `validate-data/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `visitor-intro/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+- `vm-mcp/` [not refreshed; prior source: `38c64332b840b2bbe1c07e53aeee8996644548e9`]
+
+## Lifecycle propagation usage
+
+Read `agent-instantiation/SKILL.md` for a0/a0ucns and the a0p legacy copy;
+read `a0p-instancing/SKILL.md` for a0-betatest backend. Load `wdll/SKILL.md`
+when consequential spawned work needs its scope or completion proof resolved.
+These are source-owned instructions, not evidence that EPAC deploys either runtime.
+
+To reproduce this bounded refresh from skill-lib commit
+`321e721fb660e0fea5bb8c37d943611963039a57`, run:
+
+```sh
+python tools/propagate_skills.py /path/to/epac --skills agent-instantiation
+python tools/propagate_skills.py /path/to/epac --skills agent-instantiation --apply
+```
+
+The canonical dependency closure is `wdll`, `a0p-instancing`, and
+`agent-instantiation`; no other skill refresh is implied. Full-library drift and
+older MSDMD runtime/projection work remain separate. Preserve release locks,
+scientific pins, sealed results, and accepted release-consumption evidence.
+
+The scoped work graph is `docs/work-graphs/lifecycle-skills-321e721.json`.
+The snapshot statements in `docs/PROVENANCE.md` and `docs/work-graph.json`
+remain at their qualification-time identities because sealed join-term evidence
+binds those exact bytes. This README and the scoped graph describe the current
+lifecycle installation without rewriting that historical evidence.

.agents/skills/a0p-instancing/SKILL.md

@@ -1,33 +1,10 @@
 ---
 name: a0p-instancing
-description: Methodology for instancing agents in a0-betatest (the a0p research instrument), whose model diverges from canonical a0. Load this when adding or changing an AgentInstance / CharacterSheet CRUD path, a per-instance native ZFAE weight bank or its training/distillation loop, a ZFAE inference mode, a sentinel evaluation or pending-override gate, a per-agent safetensors checkpoint, or volatile sub-context memory — anywhere under a0-betatest `backend/`. Use it before writing code that creates, addresses, trains, runs, governs, or persists an a0p agent, so the code follows a0p's per-user CRUD + native-ZFAE + sentinel model instead of a0's spawn/fork/merge model. For canonical a0 and its mirror a0ucns, use `agent-instantiation` instead — a0p does NOT have `sub_agent_spawn`, a spawn executor, or `InstanceMerge`.
+description: Agent instancing in a0-betatest backend (a0p). Load this when creating, addressing, training, running, governing, persisting, or changing AgentInstance/CharacterSheet CRUD, native ZFAE weight banks, teacher distillation, inference modes, sentinel evaluation, pending overrides, per-agent safetensors checkpoints, Mongo metadata, FIQ events, or volatile sub-context memory. Do not load for ordinary LLM-prompt work that creates no instance. For a0, a0ucns, or a0-betatest/_legacy_a0 use agent-instantiation/SKILL.md.
 ---
 
 # a0p-instancing — How a0-betatest brings agents into being
 
-a0-betatest (the **a0p** research instrument) replaced a0's single-persistent
-+ fork model with a **multi-agent, per-user model**: an agent is a persistent
-CRUD entity (`AgentInstance`) bound to an editable `CharacterSheet`, and each
-instance **owns its own trained native ZFAE weight bank**. The LLM is optional
-energy; once an instance's native core is trained, it can run with no LLM at
-all. This skill captures that model so a coding agent extends it instead of
-reaching for a0's spawn/fork/merge machinery, which **does not exist here**.
-
-This skill is the **peer** of `agent-instantiation` (which documents a0 /
-a0ucns). They describe genuinely different architectures; do not mix them.
-
-## Load this when
-
-- Adding/editing an `AgentInstance` or `CharacterSheet` **CRUD** path.
-- Touching the **native ZFAE weight bank**, its **teacher-distillation
-  training**, or the native **inference** engine / modes.
-- Adding/changing a **sentinel** evaluation or a **pending-override** gate.
-- Persisting per-agent state (**safetensors checkpoints**, Mongo metadata,
-  **FIQ** audit events) or volatile **sub-context** memory.
-
-Do not load this for canonical a0 / a0ucns work (use `agent-instantiation`),
-or for ordinary LLM-prompt work that creates no instance.
-
 ## Scope and source boundary
 
 The **canonical source is `a0-betatest/backend/`** (the a0p package per its
@@ -54,7 +31,8 @@ reproduce them from memory. Where a mechanism is unclear or unwired, write
   **fresh per agent** at creation (`backend/agents/store.py`).
 
 There is **one weight bank per agent**, not a shared singleton engine. Never
-share or fork a bank between instances.
+share or fork a bank between instances. A trained native core can run without
+an LLM; the inference mode governs optional LLM energy.
 
 ## Instancing sequence (dependency order — follow it top to bottom)
 
@@ -82,7 +60,7 @@ share or fork a bank between instances.
    or both produce/judge the reply. `zfae_native` refuses LLM fallback.
 6. **Evaluate sentinels and honor the override halt** (`sentinel_eval.py`,
    `sentinels.py`, `overrides.py`). Resolve per-agent sentinel modes/weights,
-   compute the `Verdict13`; on a blocking flag, create a `PendingOverride`
+   compute the `Verdict13` from 13 sentinels (S1–S13); on a blocking flag, create a `PendingOverride`
    and **halt** until approved. Do not bypass the halt.
 7. **Checkpoint to safetensors, not a DB blob.** Persist the bank to
    `storage/agents/{id}/zfae_core.safetensors` (+ `…meta.json`: digest,

.agents/skills/agent-instantiation/SKILL.md

@@ -1,47 +1,25 @@
 ---
 name: agent-instantiation
-description: Methodology for instantiating, forking, running, merging, and retiring agents in the a0 platform and its near-identical mirror a0ucns. Load this when adding or changing a sub-agent spawn path, a PCNA instance fork/merge, an agent definition or naming scheme, spawn caps or approval gating, an agent run/log table, a heartbeat-driven agent task, or a checkpoint of agent state. Use it before writing any code that creates, addresses, schedules, or tears down an agent or sub-agent, so the new code follows the platform's existing lifecycle, fork/merge, identity, and gating contracts rather than inventing a parallel one. NOTE — a0-betatest (a0p) has diverged to a different per-user CRUD + native-ZFAE instancing model — this skill's spawn/fork/merge sequence does NOT apply there; see "a0-betatest divergence".
+description: Agent lifecycle for a0 and its near-identical mirror a0ucns. Load this when creating, addressing, scheduling, forking, running, merging, or retiring agents; changing sub-agent spawn paths, PCNA instance fork/merge/converge, agent definitions or naming, spawn caps, spawn executor or approval gating, agent_runs or agent_logs, heartbeat tasks, or checkpoints. Do not load for ordinary LLM-call/inference-prompt work that does not create or reshape an instance. For a0-betatest backend use a0p-instancing/SKILL.md; its _legacy_a0 reference copy follows this skill.
 ---
 
 # agent-instantiation — How a0 brings agents into being
 
-a0 runs **one persistent agent** (ZFAE) and lets it **fork sub-agents**
-(`a0(model)zeta{n}`) that run in parallel and **merge back**. An agent is
-not an LLM call: the LLM is an interchangeable *energy provider*, while the
-agent *is* a `PCNAEngine` instance (six prime-indexed rings) plus its
-identity, memory, run record, and checkpoint. This skill captures the
-methodology a coding agent must follow so a new spawn path, ring fork, merge
-mode, or agent definition plugs into the existing lifecycle instead of
-growing a second, divergent one.
-
-## Load this when
-
-- Adding or editing a sub-agent **spawn** path, or any code that creates an
-  `agent_runs` row.
-- Adding a **fork**, **merge**, or **converge** of a `PCNAEngine` instance.
-- Defining a **new agent** (its name/symbol/slot/directives/tools) or
-  changing the **naming** scheme.
-- Touching **spawn caps**, **approval gating**, or the **spawn executor**.
-- Persisting agent state (**checkpoints**, **run logs**) or scheduling an
-  agent task on the **heartbeat**.
-
-Do not load this for ordinary LLM-call/inference-prompt work that does not
-create or reshape an agent instance.
-
 ## Scope and source boundary
 
 The **canonical source is `a0`** (`python/engine/`, `python/services/`,
-`python/agents/`, `shared/schema.ts`). `a0ucns` is a near-identical mirror and
-follows this skill verbatim. **`a0-betatest` (a0p) has diverged** to a
-different instancing model and does **not** follow the sequence below — see
-"a0-betatest divergence" before touching that repo.
+`python/agents/`, `shared/schema.ts`). For the a0p architecture and its
+exceptions, load [a0p-instancing](../a0p-instancing/SKILL.md).
 This is **repo-specific runtime doctrine**, not org-universal math: it
 transfers no UCNS / PCNA / PCTA theorem status (see each repo's boundary
 notes). Cite the a0 files below; if a mechanism is not in them, mark it
 `hmmm` rather than inventing it.
 
 ## The model: persistent agent vs sub-agents
 
+An agent is a `PCNAEngine` instance (six prime-indexed rings) plus identity,
+memory, run record, and checkpoint. An LLM call supplies interchangeable energy.
+
 - **Persistent agent (ZFAE).** Defined as a plain dict in
   `python/agents/zfae.py` (`ZFAE_AGENT_DEF`: `name`, `symbol`, `slot`,
   `directives`, `tools`, `sentinel_seed_indices`, `is_persistent`). Its
@@ -80,35 +58,36 @@ stream lives in `agent_logs`. Keep these three in agreement.
    `get_pcna()` at lifespan startup; `await pcna.load_checkpoint()` to
    restore learned ring state; ensure its row in the agent/instance table.
    Do not construct a second primary.
-3. **Spawn only through the `sub_agent_spawn` tool**
+3. **Bind consequential spawned work to WDLL.** When the child receives a nontrivial objective whose target state, scope, or proof is not already explicit, apply `wdll/SKILL.md` before execution so the spawn owns a falsifiable state transition rather than a broad activity prompt.
+4. **Spawn only through the `sub_agent_spawn` tool**
    (`python/services/tools/sub_agent_spawn.py`). Never `INSERT` an
    `agent_runs` row by hand. The tool checks spawn caps, derives
    `root_run_id`/`depth` from the parent run scope, inserts the row
    `status='running'`, and returns `{ok, agent_id, run_id}` immediately
    (spawn is non-blocking).
-4. **Fork the engine via `InstanceMerge.fork(parent)`**
+5. **Fork the engine via `InstanceMerge.fork(parent)`**
    (`python/engine/merge.py`) — returns `(child, meta)`. The child gets
    independent tensors with small Gaussian noise (a0: σ≈0.02 on Φ/Ψ/Ω,
    ≈0.01 on Θ; Memory-L copied deterministically). Register it in
    `_sub_agents` with `parent_id` + `run_id`. Never share tensor references
    between instances.
-5. **Execute via the spawn executor, not inline.** The background loop in
+6. **Execute via the spawn executor, not inline.** The background loop in
    `python/services/spawn_executor.py` claims one `running` row atomically
    (`SELECT … FOR UPDATE SKIP LOCKED` in `spawn_db.py`), resolves the
    provider, runs one turn, emits to `agent_logs`, and sets the terminal
    status (with the row's retry policy on transient errors). Do not call the
    model directly from the spawn path.
-6. **Merge with `InstanceMerge.absorb(parent, child)`** when the child's
+7. **Merge with `InstanceMerge.absorb(parent, child)`** when the child's
    work is done — federated averaging blends the rings (a0: donor α≈0.15 on
    Φ/Ψ/Ω, ≈0.8 on Memory-L). Then unregister the child, mark its row
    `merged`, and archive its log stream. Use `fork`/`absorb` for the
    parent⇄child path; `converge(a, b, α)` only for two live peers that both
    continue.
-7. **Persist on a cadence, validate on restore.** Save ring tensors to the
+8. **Persist on a cadence, validate on restore.** Save ring tensors to the
    checkpoint store (a0: base64 in `system_toggles`) from a **heartbeat
    task**, not ad hoc; on load, validate every ring's shape and assign
    nothing if any mismatches (all-or-nothing restore).
-8. **Gate every mutation.** Manual spawn/merge routes call
+9. **Gate every mutation.** Manual spawn/merge routes call
    `require_admin(request)` (or are listed in the gating allowlist with a
    justification). Tools that cause side effects honor the approval scope
    (`get_approval_scope_user_id()`); spawn caps and parent run scope ride
@@ -149,10 +128,6 @@ username( a0( <energy / inference provider> ) <auditor / teacher / …> )
 | `a0(zfae)` | native ZFAE engine | — |
 | `a0(zfae)gpt 5.5` | native ZFAE engine | gpt 5.5 (teacher / auditor) |
 
-Energy is inside the parens, auditor is outside, the user wraps the whole
-thing. `zfae` *inside* the parens means native inference is the energy — it
-is not an auditor.
-
 > Reconciliation note (`hmmm`): a0's current code emits a different,
 > pre-nomenclature form — `compose_name(...)` → `a0({model})zfae` and
 > `sub_agent_name(index, ...)` → `a0({model})zeta{index}` — where the trailing
@@ -199,39 +174,6 @@ is not an auditor.
 - Skipping `require_admin` / approval-scope checks on a new spawn or merge
   surface.
 
-## a0-betatest divergence (does not follow this skill)
-
-`a0-betatest` (a0p) replaced the a0 model wholesale. Do **not** apply the
-spawn/fork/merge sequence there; its instancing is:
-
-- **An instance is a per-user CRUD entity, not a forked singleton.**
-  `AgentInstance` (UUID + editable `CharacterSheet`) created/read/updated/
-  archived via routes — `backend/agents/{schema.py,store.py,routes.py}`.
-  There is no single persistent `PCNAEngine`; each agent is its own entity.
-- **Each instance owns a native ZFAE weight bank**, not shared ring tensors:
-  three 157-seed cores `[157,53,7,7]` (1,223,187 scalars), `A0ZFAEWeightBank`
-  in `backend/interdependent_lib/zfae/weights.py`; the native engine refuses
-  LLM fallback in `zfae_native` mode and trains by teacher distillation.
-- **No `sub_agent_spawn`, spawn executor, or `InstanceMerge`.** The only
-  "spawn/merge" is volatile in-memory sub-context scoping —
-  `MemoryCore.spawn_sub` / `merge_sub` in
-  `backend/interdependent_lib/pcna/memory_core.py`. No ring/weight forking,
-  no federated averaging, no `agent_runs` state machine.
-- **Identity is a UUID + character-sheet name**, per `user_id` — not the
-  `a0(model)zfae` / `zeta{n}` naming convention.
-- **Persistence is filesystem + Mongo + FIQ**, not `system_toggles` /
-  `agent_logs`: per-agent `storage/agents/{id}/zfae_core.safetensors` (+ meta
-  JSON), agent metadata in the Mongo `agent_instances` collection, and a
-  hash-chained **FIQ audit log** for events.
-- **Gating is sentinel + override**, not spawn caps + ContextVars: 13
-  sentinels (S1–S13) with per-agent modes/weights and a pending-override
-  halt gate, plus `MODULE_BUILD`/`BOUNDARIES`/`CAPABILITIES`/`RATIOS`
-  enforced by the vendored `a0p_skills` runners.
-
-`a0-betatest/_legacy_a0/` is a reference copy of canonical a0 (which *does*
-follow this skill). If a0-betatest's per-instance native-ZFAE model needs its
-own doctrine, it belongs in a **separate** skill, not by stretching this one.
-
 ## hmmm
 
 - a0's `fork()` seeds its RNG from `time.time()`; rapid successive forks can