mishkan-harness 0.2.5 → 0.2.7

package/docs/usage/10-observability.md

@@ -24,7 +24,7 @@ Auto-installed during `npx mishkan-harness install` (phase 7) — the installer
 prompts once. Or run standalone any time:
 
 ```bash
-npx mishkan-harness observability
+npx mishkan-harness observability install
 ```
 
 Requirements: `uv` (https://astral.sh/uv) and Python 3.11+. If `uv` is
@@ -90,7 +90,7 @@ mishkan-watchd stop                # SIGTERM via the PID file
 | `1` | **Live** | "What is happening *right now*?" Active agents (with `alias · role` annotation), workflows in-flight, current worktrees, Cognee + Graphify + MCP rollup, rolling feed of every event. Default tab on launch. |
 | `2` | **Agents** | "What did each agent do?" Sessions tree (left, project paths decoded) × agent history `DataTable` (centre) × errors panel (right). Phantom sessions filtered out at the daemon `_confirmed_alive` gate. |
 | `3` | **Workflows** | "What did each dynamic workflow run, and what's *available* to run?" Recent runs (top) + static catalogue parsed from each script's `meta` block (description, when-to-use, phases). Click an entry for detail. |
-| `4` | **Knowledge** | "Are my stores up and growing?" Three cards (work + curated + graphify, real-time counts) × recent ops `DataTable` × MCP server status table. |
+| `4` | **Knowledge** | "Are my stores up and growing?" Three cards (`cognee-memory` + `cognee-curated` + graphify, real-time counts) × recent ops `DataTable` × MCP server status table. |
 | `5` | **Activity** | Unified, filterable event stream. Regex filter + type/agent selects. Errors and blocked permissions break the visual rhythm with a separator line. |
 | `6` | **Org-Ref** | Read-only browser of the 45-agent org from `org.json`. Tree by team, click a group → mission / charter / Hebrew name; click an agent → role, source, description. |
 | `7` | **Usage** | Harness-wide tokens in/out/cached, cost, context window estimate, request counts. Per-session table sorted by token volume. Detail panel with per-agent attribution + top tools. |
@@ -126,7 +126,7 @@ Six daemon sources, each independent and fail-open:
 | `session_discover` | 10 s | active Claude Code sessions (mtime < 60 s) |
 | `worktree_poll` | 5 s | `git worktree list --porcelain` per known project |
 | `mcp_probe` | 60 s | discover MCPs from `~/.claude.json` + `.mcp.json` + `mcp-needs-auth-cache.json`; probe each by HTTP/TCP |
-| `cognee_poll` | 30 s | HTTP probe of cognee work + curated + cypher node count |
+| `cognee_poll` | 30 s | HTTP probe of `cognee-memory` (`:7777`) + `cognee-curated` (`:7730`) + cypher node counts |
 | `session_tail` + `subagent_tail` | 3 s | inter-agent + compaction events from main session JSONL; subagent tool calls from nested `subagents/agent-*.jsonl` |
 
 ## Keybindings

package/docs/usage/11-graphify.md

@@ -1,27 +1,47 @@
 # 11 — Graphify (code-structure graph)
 
-> Goal: use Graphify as the **structural** layer of the MISHKAN
-> three-store knowledge stack (Graphify · Cognee work · Cognee curated)
-> so that "who calls X" / "what depends on Y" questions cost ~1.8k
-> tokens instead of lifting the whole repo into context.
+> Goal: use Graphify as the **code-structure** surface of the MISHKAN
+> four-surface knowledge model — 1 code-structure (Graphify) + 3 cognee
+> stores (work · memory · curated) — so that "who calls X" / "what
+> depends on Y" questions cost ~1.8k tokens instead of lifting the whole
+> repo into context.
 
 ## What it is
 
 [Graphify](https://github.com/safishamsi/graphify) is a tree-sitter
 based code-graph extractor with a query path that traverses the graph
 to answer structural questions. MIT, Python, `uv tool install`-able.
-Per D-008 of the MISHKAN decision log, Graphify is the third store of
-the knowledge stack:
 
-| Store | Question it answers | Source |
-|---|---|---|
-| **Graphify** | *How is the code structured?* | tree-sitter AST + optional LLM enrichment, deterministic, re-derivable |
-| **Cognee work** | *Why does this code exist and what did we decide?* | agent-ingested ADRs, runbooks, resolved research |
-| **Cognee curated** | *What did we learn on other projects?* | seeded, read-only from projects |
+MISHKAN has **four knowledge surfaces** — one code-structure surface
+(Graphify) and three cognee stores — each owning one epistemic question
+with no overlap of write authority:
+
+| Surface | Question it answers | Source | D-ref |
+|---|---|---|---|
+| **Graphify** (per-project, `graphify-out/`) | *How is the code structured?* | tree-sitter AST + optional LLM enrichment, deterministic, re-derivable | D-008 |
+| **Cognee work** (per-project Ladybug, own port) | *Why does this code exist and what did we decide?* | agent-ingested ADRs, runbooks, resolved research | D-008, D-012 |
+| **cognee-memory** (`:7777`, shared) | *What does the session remember across all projects?* | `claude_code_memory` dataset; shared, not re-derivable from docs | D-012 |
+| **cognee-curated** (`:7730`, shared) | *What did we learn on other projects?* | promoted cross-project reference library, read-only from projects | D-007 |
+
+The **D-008 framing** ("three stores: Graphify · Cognee work · Cognee
+curated") reflects the stack as it stood before D-012. D-012 added the
+`cognee-memory` pillar (repurposing the old shared `:7777` Neo4j box
+for session memory only), bringing the total to four. The two framings
+are not contradictory — they count different cuts: D-008 defines the
+write-discipline split; D-012 adds a fourth surface within the cognee
+tier. The crisp routing test remains:
+**structure → Graphify, project semantics → Cognee work.**
+
+```mermaid
+flowchart TD
+    Q{"What kind of question?"}
+    Q -->|"structure: who calls X, what depends on Y"| GFY[("Graphify<br/>code-structure graph · per-project")]
+    Q -->|"why we decided X (this project)"| WORK[("cognee work<br/>per-project · isolated")]
+    Q -->|"what the session remembers (all projects)"| MEM[("cognee-memory :7777<br/>shared session memory")]
+    Q -->|"what we learned on other projects"| CUR[("cognee-curated :7730<br/>shared reference · read-only")]
+```
 
-These are non-overlapping by design: one writer per store, one
-question type per store, no semantic mixing. The crisp test:
-**structure → Graphify, semantics → Cognee work.**
+*Four surfaces, one epistemic question each — no overlap of write authority.*
 
 ## Install
 
@@ -252,13 +272,13 @@ opted out of the structural layer.
 
 ## See also
 
-- [D-008](../design/MISHKAN_decisions.md#d-008) — three-store knowledge
-  epistemology.
+- [D-008](../design/MISHKAN_decisions.md#d-008) — three-layer knowledge
+  epistemology (Graphify + Cognee work + Cognee curated; pre-D-012 framing).
 - [D-009](../design/MISHKAN_decisions.md#d-009) — graph-first PreToolUse
   advisory hook (Phase 1 telemetry-only, Phase 2 advisory injection —
   both shipped v0.2.3).
 - [POC report](../research/graphify-token-saving-poc.md) — verified
   88.1× reduction on the MISHKAN harness, 2026-06-07.
-- [Memory layer](./04-memory-layer.md) — the Cognee work and curated
-  stores that complete the three-store stack.
+- [Memory layer](./04-memory-layer.md) — the three cognee stores (work,
+  memory, curated) that complete the four-surface knowledge model.
 - Upstream — https://github.com/safishamsi/graphify.

package/docs/usage/12-cli.md

@@ -0,0 +1,87 @@
+# 12 — The `mishkan` CLI
+
+> One control surface for the harness and its knowledge stack. ADR **D-015**.
+
+Every command is **`mishkan <object> <verb>`** — it names *what it operates on*.
+The one exception is self-management (`install` / `uninstall` / `status`), where the
+object is the tool itself (so `npx mishkan-harness install`, not the doubled
+`… harness install`).
+
+> **Who runs these:** you do. The CLI executes because a *human* invokes it. Agents
+> never get this bin (rule 5 / D-005), and the observability TUI only *surfaces*
+> commands to copy — it never runs them.
+
+## Manage the harness
+
+| Command | Does |
+|---|---|
+| `mishkan install` | Install / refresh into `~/.claude` (idempotent) |
+| `mishkan uninstall [--purge]` | Remove it (`--purge` also drops `y4nn-standards.md`) |
+| `mishkan status` | Install state **+ live knowledge-stack health** in one view |
+
+## Knowledge
+
+| Command | Does |
+|---|---|
+| `mishkan knowledge configure` | Wizard: LLM provider + cognee secrets → writes `.env` + `ACCESS.txt` |
+| `mishkan knowledge ingest [--tagged-only] [--dataset=X] [paths…]` | Add docs to **this project's** store (`add → cognify → memify`) |
+| `mishkan knowledge curate` | Review + approve research-found resources into the **shared curated library** (D-016). Walks the candidate queue Baruch fills; on approval, an *additive* (no-prune, dedup) write. Stateful — you run it. |
+| `mishkan knowledge reset` | **Destructive** — full reset of the knowledge layer to the stable baseline: wipes every work store (container + volume), prunes `cognee-memory`, re-seeds `cognee-curated` from the canonical YAML. Type-to-confirm. Stateful — you run it. Work stores recreate on the next `/mishkan-init`. |
+| `mishkan knowledge-stack up [--build]` | Bring the shared infra up (memory `:7777` + curated `:7730` + ollama/pg). **Guided:** preflights config, names any gap, stops — never a cryptic docker error. `--build` only for the first image build. |
+| `mishkan knowledge-stack down` | Stop it — containers down, volumes/data survive (confirms) |
+| `mishkan knowledge-stack restart` | down + up |
+| `mishkan knowledge-stack status` | Detailed per-container health |
+| `mishkan project-work-store [<slug>] up` | Provision this project's own isolated store (slug → current dir) |
+| `mishkan project-work-store [<slug>] down` | Stop it — data volume kept |
+| `mishkan project-work-store [<slug>] reset` | Wipe it — removes container **and** data volume (confirms) |
+
+**`knowledge-stack` vs `project-work-store`** is infra-lifecycle (`up` a shared stack)
+vs data-lifecycle (`reset`/wipe a per-project store) — named differently on purpose.
+
+## Inspect / observe
+
+| Command | Does |
+|---|---|
+| `mishkan code-graph status \| open \| scan` | The project's code graph (Graphify) |
+| `mishkan observability install \| open` | The live monitor — install the daemon+TUI, or open it |
+| `mishkan org show [--json]` | The 45-agent organisation reference |
+| `mishkan model show \| set \| reset` | Re-tier which Claude model an agent runs on (D-017) |
+
+## Re-tier the agent fleet — `mishkan model`
+
+Every MISHKAN agent runs on a Claude tier (D-002). The shipped defaults live in
+`model-routing.yaml`; **your** overrides live in `model-routing.local.yaml`, an overlay
+the installer never clobbers — so re-tiering survives `mishkan install`, and you never
+edit 45 frontmatter files.
+
+| Command | Does |
+|---|---|
+| `mishkan model show` | Effective tier per agent (default + your overrides), marking overrides and flagging any **dormant** tier |
+| `mishkan model set <agent\|team\|all> <tier>` | Override a tier. `<tier>` ∈ `opus \| sonnet \| haiku \| fable`. `<team>` is an org id (`mishmar`, `migdal`, …); `all` = every agent |
+| `mishkan model reset [<agent\|team\|all>]` | Drop override(s); no argument clears them all (back to shipped defaults) |
+
+Changes take effect on the **next delegation** — the `model-route.py` hook reads the
+overlay live; no reinstall needed. Use it for cost (drop a tier), availability (a tier
+can be suspended — see the **fable** note below), or preference.
+
+> **`fable` is dormant.** Claude Fable 5 was suspended for all customers on 2026-06-12
+> (export-control directive). `mishkan model set … fable` warns and confirms; agents
+> routed to it will fail to spawn until access is restored. The tier value is kept so
+> re-enabling is one command if it returns.
+
+The TUI binary is `mishkan-watch`; `mishkan-watchd start|stop|status` is the manual
+daemon control. In the TUI, **`c`** copies the fix command for a down store to your
+clipboard (it shows the command — it never runs it).
+
+## Project init composes the verbs
+
+`/mishkan-init` runs `knowledge-stack up` (ensure the shared infra, confirm-if-down)
+then `project-work-store up` (this project's store) — the same guided/preflight path,
+so a fresh project comes up working without you knowing the topology.
+
+## Deprecated aliases
+
+The pre-D-015 flat names still work (hidden, undocumented) so nothing breaks mid-
+migration: `configure-knowledge` → `knowledge configure`, `ingest` → `knowledge
+ingest`, `observability` → `observability install`, `watch` → `observability open`,
+`org` → `org show`. Prefer the object-first forms.

package/docs/usage/12-skill-discovery.md

@@ -94,6 +94,17 @@ Output shape (abbreviated):
 over padding `must_load` when scores are close — the cost of loading a
 wrong skill is higher than the cost of skipping a marginal one.
 
+```mermaid
+flowchart LR
+    IDX[("skill index<br/>all installed skills")] --> R{"router<br/>score vs thresholds"}
+    R -->|"≥ 4.0 + strong trigger"| ML["must_load ≤ 3<br/>load into context"]
+    R -->|"mid-band"| SC["should_consider ≤ 5<br/>skim · load if it intersects"]
+    R -->|"low · same category"| AD["adjacent ≤ 5<br/>awareness only"]
+```
+
+*Indexer → router → three buckets (hard cap 13); bias toward enriching `should_consider`
+over padding `must_load`.*
+
 ## Trust asymmetry
 
 Every entry carries an `origin` field. Non-`mishkan` entries (origin =

package/docs/usage/README.md

@@ -6,31 +6,46 @@
 
 A single Claude Code session, turned into a 45-agent software-engineering
 organisation with deterministic constraints (hooks, rules, schemas), an
-asymmetric AI-vs-human delegation boundary, and a two-store knowledge graph
-that accumulates as you work.
+asymmetric AI-vs-human delegation boundary, and four knowledge surfaces
+that accumulate as you work: one code-structure surface (Graphify) plus three
+cognee stores (per-project work · session memory · curated reference).
 
 ## In five minutes
 
+```mermaid
+flowchart TD
+    You(["You — talk"]) --> MS[["MAIN SESSION — leadership<br/>Nehemiah + Bezalel (via CLAUDE.md)<br/>the ONE orchestrator — no nested delegation"]]
+    MS -->|Task| TL["Team Lead / Specialist"]
+    MS -->|Task| OPS["aiobi-ops / other project agents"]
+    MS -->|Task| RP["research pipeline"]
+    MS -. reads / writes .-> WORK[("cognee work · per-project<br/>mishkan-work-{slug} · Ladybug, own port+volume")]
+    MS -. reads / writes .-> MEM[("cognee-memory :7777<br/>per-client session memory · shared")]
+    MS -. reads / writes .-> CUR[("cognee-curated :7730<br/>reference library · shared, read-mostly")]
 ```
-You ──talk──▶  MAIN SESSION  = leadership (Nehemiah/Bezalel via CLAUDE.md)
-                   │            ← the ONE orchestrator (no nested delegation)
-                   ├─Task→ Team Lead / Specialist            ┐ siblings,
-                   ├─Task→ aiobi-ops or other project agents │ one level deep
-                   └─Task→ research pipeline                 ┘
-                                                                ↓
-                    ┌──────────────────────────┐    ┌──────────────────────┐
-                    │ cognee WORK   :7777      │    │ cognee CURATED :7730 │
-                    │ project knowledge        │    │ reference library    │
-                    │ + per-client memory      │    │ (read-mostly, shared)│
-                    └──────────────────────────┘    └──────────────────────┘
-```
+
+*The main session is the one orchestrator — it delegates one level deep (siblings) and
+reads/writes three Cognee stores. (Diagrams render on GitHub.)*
 
 - **Main session is leadership.** It loads MISHKAN identity from
   `~/.claude/CLAUDE.md` and routes work one level deep.
 - **45 agents** across **6 teams** + **2 orchestrators** + a **6-stage research
   pipeline**.
-- **Cognee** is the memory layer: two physically-isolated stores, with
-  `cognify → memify` (extraction → enrichment) and `search` exposed via MCP.
+- **Four knowledge surfaces** total — 1 code-structure + 3 cognee stores:
+  - **Graphify** (per-project, `graphify-out/`) — code structure: call graphs,
+    dependents, blast-radius. AST-derived, deterministic, re-derivable. D-008.
+    Not installed by `mishkan install`; run `uv tool install "graphifyy>=0.8.33"`.
+  - **Cognee work** (per-project Ladybug, own port) — project semantic memory:
+    decisions, runbooks, ADRs. D-012.
+  - **cognee-memory** (`:7777`, shared) — session memory shared across all
+    projects; `claude_code_memory` dataset only.
+  - **cognee-curated** (`:7730`, shared) — cross-project reference library;
+    read-mostly, promoted at `/sprint-close`. D-007.
+
+  The D-008 framing ("three stores") predates D-012 and counts Graphify + the two
+  cognee stores as they existed then. D-012 added the `cognee-memory` pillar,
+  making the total four. The `cognify → memify` (extraction → enrichment) and
+  `search` operations are exposed via MCP for the three cognee stores. See
+  D-007 + D-008 + D-012.
 - **Selective ingest**: docs enter the work graph only when tagged
   (`mishkan: ingest`) or explicitly invoked. No bulk-ingest, no PII bleed.
 
@@ -41,14 +56,15 @@ You ──talk──▶  MAIN SESSION  = leadership (Nehemiah/Bezalel via CLAUDE
 | 01 | [Installation](./01-installation.md) | Prerequisites, `npx mishkan-harness install`, layout, uninstall |
 | 02 | [Project initialisation](./02-project-init.md) | `/mishkan-init` flow, scope choices, brownfield handling |
 | 03 | [Orchestration](./03-orchestration.md) | Main-session-as-orchestrator, model routing, skills on-demand |
-| 04 | [Memory layer (cognee)](./04-memory-layer.md) | Work + curated stores, `cognify`/`memify`/`search`, UIs |
+| 04 | [Memory layer (cognee)](./04-memory-layer.md) | Per-project work stores + `cognee-memory` (`:7777`) + `cognee-curated` (`:7730`), `cognify`/`memify`/`search`, UIs |
 | 05 | [Selective ingest](./05-selective-ingest.md) | `mishkan-ingest`, frontmatter tagging, memory-is-opt-in |
 | 06 | [LLM provider profiles](./06-llm-providers.md) | Gemini/NVIDIA/Ollama/OpenAI/Anthropic, rate vs daily caps |
 | 07 | [Troubleshooting](./07-troubleshooting.md) | Real gotchas + fixes from the build |
 | 08 | [Glossary](./08-glossary.md) | 45-agent roster (alias → role → team), key terms |
-| 09 | [Dynamic Workflows](./09-workflows.md) | 10 org-level + 8 team-level workflows, ADR D-010 portfolio discipline |
+| 09 | [Dynamic Workflows](./09-workflows.md) | 10 org-level + 10 team-level workflows, ADR D-010 portfolio discipline |
 | 10 | [Observability](./10-observability.md) | Cross-session daemon + Textual TUI; 8 tabs (Live · Agents · Workflows · Knowledge · Activity · Org-Ref · Usage · Skills), project filter (`p`) |
 | 11 | [Graphify](./11-graphify.md) | Code-structure graph; queries at ~1.8k tokens (88.1× reduction, POC-verified); D-008 + D-009 |
+| 12 | [The `mishkan` CLI](./12-cli.md) | One control surface — semantic `<object> <verb>`: `knowledge-stack` / `project-work-store` / `knowledge` (configure · ingest · curate · reset) / `model` (re-tier agents) / `observability` / `org`; D-015 · D-016 · D-017 |
 | 12 | [Skill discovery](./12-skill-discovery.md) | Universal indexer + 3-bucket router across MISHKAN, user, plugin, and project skills; D-011 |
 
 ## Where to start
@@ -62,7 +78,7 @@ You ──talk──▶  MAIN SESSION  = leadership (Nehemiah/Bezalel via CLAUDE
 ## Authoritative references this documentation builds on
 
 - [`docs/design/MISHKAN_harness_design.md`](../design/MISHKAN_harness_design.md) — the 5-layer architecture and rationale.
-- [`docs/design/MISHKAN_decisions.md`](../design/MISHKAN_decisions.md) — D-001…D-011 with rationale.
+- [`docs/design/MISHKAN_decisions.md`](../design/MISHKAN_decisions.md) — D-001…D-012 with rationale.
 - [`docs/design/MISHKAN_agent_aliases.md`](../design/MISHKAN_agent_aliases.md) — the biblical roster.
 - [`docs/design/MISHKAN_ontology.md`](../design/MISHKAN_ontology.md) — cognee entity + relationship types.
 - [`docs/design/MISHKAN_token_optimisation.md`](../design/MISHKAN_token_optimisation.md) — context economy.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mishkan-harness",
-  "version": "0.2.5",
+  "version": "0.2.7",
   "description": "MISHKAN — a personal advanced SWE R&D harness for Claude Code: 45 biblically-named agents across six teams, deterministic rules + hooks, a shared research pipeline, dependency/supply-chain vetting, and a Cognee-backed knowledge graph. Installs into ~/.claude.",
   "type": "module",
   "bin": {

package/payload/mishkan/AGENT_SPEC.md

@@ -97,7 +97,9 @@ load-bearing rules. It is the first line of defence; the security hook
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 ```
 
 Role-specific constraints (e.g. "pnpm only" for Salma, "OpenAPI 3.1 first"

package/payload/mishkan/agents/ahikam.md

@@ -46,7 +46,9 @@ You collect and assemble Panim's milestone report.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 No `/plan` (collect-only role).
 

package/payload/mishkan/agents/aholiab.md

@@ -56,7 +56,9 @@ handed off, the design decisions and their rationale, and what is out of scope.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 Approval gate via `/plan`.
 

package/payload/mishkan/agents/asaph.md

@@ -62,7 +62,9 @@ finding:
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 WCAG 2.2 AA minimum.
 

package/payload/mishkan/agents/baruch.md

@@ -34,6 +34,11 @@ You are the terminal stage. You record the research outcome faithfully.
 - On `outcome: resolved` with cross-harness blast radius, write a Cognee node
   (ResearchOutput or CaseNode per ontology) and set `knowledge_graph_write: true`
   and `cognee_node_id`.
+- When Shemaiah set a `curated_promotion_candidate` (D-016), copy it verbatim into
+  the research-log AND append the candidate object to the engineer queue
+  `~/.claude/mishkan/curated-candidates.jsonl` (one JSON object per line). You
+  **queue** a candidate — you have no curated-write tool; the engineer approves it
+  via `mishkan knowledge curate`. Never auto-promote, never `docker exec`.
 
 ## Output discipline — non-negotiable
 
@@ -78,7 +83,9 @@ to prevent.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 ## Contract eval
 

package/payload/mishkan/agents/benaiah.md

@@ -64,7 +64,9 @@ categories to be covered, assets in scope, and trust boundaries.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 Two root causes on non-trivial failures.
 

package/payload/mishkan/agents/bezalel.md

@@ -71,7 +71,9 @@ Do not proceed until Y4NN approves. The approved plan is the scope contract.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 Approval gate on consequential decisions via `/plan`.
 

package/payload/mishkan/agents/caleb.md

@@ -64,7 +64,9 @@ coverage: <which sub-questions were answered, which were not>
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 ---
 

package/payload/mishkan/agents/deborah.md

@@ -51,7 +51,9 @@ emotional response, trust.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 No fabricated research — cite the heuristic or study.
 

package/payload/mishkan/agents/elasah.md

@@ -46,7 +46,9 @@ You collect and assemble Chosheb's milestone report.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 No `/plan` (collect-only role).
 

package/payload/mishkan/agents/eliashib.md

@@ -56,7 +56,9 @@ no deploy proceeds past an open critical finding.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 No `:latest` tags. SOPS for secrets. Hardening overlay on recreate. Approval gate via `/plan`.
 

package/payload/mishkan/agents/ezra.md

@@ -59,7 +59,9 @@ curated_library_match: true|false
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 ---
 

package/payload/mishkan/agents/hanun.md

@@ -53,7 +53,9 @@ operational glue.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 Hardening overlay on every recreate. SOPS for secrets.
 

package/payload/mishkan/agents/hiram.md

@@ -57,7 +57,9 @@ component contracts, or shared visual primitives).
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 WCAG 2.2 AA on interactive components.
 

package/payload/mishkan/agents/hizkiah.md

@@ -65,7 +65,9 @@ when a task touches more than one component.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 OpenAPI contract before endpoint.
 

package/payload/mishkan/agents/huldah.md

@@ -47,7 +47,9 @@ You verify and report Sefer's milestone work.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 No `/plan` (collect-only role).
 

package/payload/mishkan/agents/huram.md

@@ -54,7 +54,9 @@ Yasad API contracts (bidirectional), and you deliver the visible product.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 Approval gate via `/plan`. pnpm only. WCAG 2.2 AA. Core Web Vitals budgets.
 

package/payload/mishkan/agents/hushai.md

@@ -50,7 +50,9 @@ you do not write or block code (that is Ira/Joab/Benaiah).
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 ---
 

package/payload/mishkan/agents/igal.md

@@ -46,7 +46,9 @@ You collect and assemble Yasad's milestone report.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 No `/plan` (collect-only role).
 

package/payload/mishkan/agents/ira.md

@@ -77,7 +77,9 @@ finding:
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 ---
 

package/payload/mishkan/agents/jahaziel.md

@@ -60,7 +60,9 @@ finding:
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 No `/plan` (evaluate against known rules).
 

package/payload/mishkan/agents/jakin.md

@@ -56,7 +56,9 @@ ready_for_formulation: true|false
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 ---
 

package/payload/mishkan/agents/jehonathan.md

@@ -51,7 +51,9 @@ You take structured graph knowledge and make it legible. You publish.
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 Diátaxis quadrant declared.
 

package/payload/mishkan/agents/jehoshaphat.md

@@ -56,7 +56,9 @@ Cognee and Team Reporter outputs and writes to `docs/` only — never to the cod
 
 Stateful operations hard stop. Sequence before implementation. Diagnose
 before fix. Durable solutions only. No scope expansion. No fabricated
-facts. English for all output.
+facts. Surface an unknown you cannot resolve up to the main session rather
+than guessing — a subagent cannot delegate onward (its Task tool is inert).
+English for all output.
 
 Diátaxis quadrant on every doc. MADR for ADRs. Keep a Changelog. No undated docs. Approval gate via `/plan`.