@event4u/agent-config 5.10.0 → 6.0.0

package/.agent-src/commands/judge/solo.md

@@ -1,95 +0,0 @@
----
-model_tier: high
-name: judge:solo
-tier: 2
-cluster: judge
-sub: solo
-skills: [subagent-orchestration]
-description: Run a standalone judge on an existing diff or code change — no implementer, no revision loop, verdict only
-suggestion:
-  eligible: false
-  rationale: "Sibling of /review-changes — eligibility routed there; keep this explicit."
-workspaces:
-  - agent-config-maintainer
-packs:
-  - meta
----
-
-# /judge solo
-## Instructions
-
-### 1. Identify the target
-
-The user invoked `/judge solo` on one of:
-
-- The current working-tree diff (`git diff`)
-- A specific commit range (`git diff A..B`)
-- A PR by number or URL
-- A single file or set of files
-
-Confirm the target in one sentence before proceeding. Do not silently
-assume the working tree.
-
-### 2. Gather the context
-
-Collect what the judge needs:
-
-- The diff in full
-- The original task, ticket, or PR description
-- Relevant acceptance criteria (tests, linter config, design doc)
-
-If any of these is missing and the judgment would be shallow without
-it, **stop** and ask.
-
-### 3. Resolve the judge model
-
-Read `.agent-settings.yml`:
-
-- `subagents.judge_model` → empty = one tier above session model
-
-Unknown alias → stop. Silence is not a fallback.
-
-### 4. Invoke the judge
-
-The judge reviews the diff against the task. Return one of:
-
-| Verdict  | Meaning                                      |
-|----------|----------------------------------------------|
-| `apply`  | Correct, complete, ready to land             |
-| `revise` | Specific issues listed — user decides next   |
-| `reject` | Fundamentally wrong approach                 |
-
-Unlike `/judge on-diff`, this command does **not** loop — no revision
-cycles. The verdict is the deliverable.
-
-### 5. Report
-
-```
-Mode:    judge (standalone)
-Target:  <diff | commit range | PR #N | files>
-Judge:   <resolved model>
-Verdict: apply | revise | reject
-Issues:  <numbered list if revise/reject>
-Next step: <what the user should do based on the verdict>
-```
-
-## Use this command when
-
-- A diff already exists and the user wants a second opinion
-- Reviewing someone else's PR before approving it
-- Double-checking a change before opening a PR
-- Stress-testing a refactor without committing to revisions
-
-## Do NOT
-
-- NEVER apply the diff — this command judges only
-- NEVER loop into a revision cycle — use `/judge on-diff` for that
-- NEVER run the judge on an empty diff — fail fast
-- NEVER silently fall back to the session model for the judge
-
-## See also
-
-- [`subagent-orchestration`](../skills/subagent-orchestration/SKILL.md)
-- [`/judge on-diff`](do-and-judge.md) — if a revision loop is wanted
-- [`/review-changes`](review-changes.md) — human-oriented self-review
-- [`role-contracts`](../../docs/guidelines/agent-infra/role-contracts.md#reviewer) — Reviewer mode output contract (Summary / Risks / Findings / Required actions / Verdict)

package/.agent-src/commands/judge/steps.md

@@ -1,93 +0,0 @@
----
-model_tier: high
-name: judge:steps
-tier: 2
-cluster: judge
-sub: steps
-skills: [subagent-orchestration, verify-completion-evidence]
-description: Execute an ordered plan step by step with a judge gate between steps — stops on first failed verdict
-suggestion:
-  eligible: false
-  rationale: "Subagent orchestration — overlaps /work and /roadmap:process-*; keep explicit."
-workspaces:
-  - agent-config-maintainer
-packs:
-  - meta
----
-
-# /judge steps
-## Instructions
-
-### 1. Confirm the plan
-
-The user invoked `/do-in-steps` pointing to a plan (agents/plans/ file,
-roadmap phase, or inline list). Confirm:
-
-- N steps, numbered
-- Each step has a clear acceptance criterion
-- Dependencies between steps are explicit
-
-If any step is vague, **stop** and ask before running anything.
-
-### 2. Resolve model pairing
-
-Same as `/do-and-judge`:
-
-- `subagents.implementer_model` → session model by default
-- `subagents.judge_model` → one tier up by default
-
-Stop if both resolve to the same model in the same context.
-
-### 3. Run the loop
-
-For each step `i` from 1 to N:
-
-1. **Implement step i** in the current session
-2. **Run the judge** on the diff for step i alone, with the plan's
-   step-i acceptance criterion as the check target
-3. **Apply or revise**:
-   - `apply` → commit step i, run targeted tests, move to step i+1
-   - `revise` → hand back the issue list, re-implement step i (max 2 retries)
-   - `reject` → **stop the whole loop**, report which step and why
-
-### 4. Stop conditions
-
-The loop exits cleanly only when **all N steps pass judgment**. Any
-other outcome is a stop:
-
-- A step rejected → stop at that step
-- Revision ceiling hit on any step → stop at that step
-- User interrupts → stop immediately
-
-On stop, report the last verified state so the user can resume.
-
-### 5. Final verification
-
-After step N passes judgment, run `verify-completion-evidence` across the
-whole changeset — targeted tests + full suite + quality pipeline.
-Never declare the plan "done" without this gate.
-
-### 6. Report
-
-```
-Mode:       do-in-steps
-Steps:      <passed>/<total>  (stopped at step <K>? yes/no)
-Implementer: <resolved model>
-Judge:      <resolved model>
-Revisions:  step-by-step count
-Evidence:   final test + quality output
-Next step:  <commit / open PR / resume from step K>
-```
-
-## Safety gates
-
-- Never skip a step because the next one looks easier
-- Never apply a revised step without the judge re-verdicting
-- Never declare complete without final `verify-completion-evidence`
-- A failing step never cascades silently — stop, report, hand back
-
-## See also
-
-- [`subagent-orchestration`](../skills/subagent-orchestration/SKILL.md)
-- [`verify-completion-evidence`](../skills/verify-completion-evidence/SKILL.md)
-- [`/do-and-judge`](do-and-judge.md) — single-change variant

package/.agent-src/commands/judge.md

@@ -1,60 +0,0 @@
----
-model_tier: medium
-name: judge
-tier: 1
-description: Judge orchestrator — routes to solo, steps, on-diff
-cluster: judge
-type: orchestrator
-suggestion:
-  eligible: true
-  trigger_description: "judge this diff, review with verdict, run an implementer→judge loop, step-by-step judged execution"
-  trigger_context: "user wants a verdict on a diff, a do-and-judge loop, or a step-gated execution"
-workspaces:
-  - agent-config-maintainer
-packs:
-  - meta
----
-
-# /judge
-
-Top-level orchestrator for the `/judge` family. Replaces 3 standalone
-commands with a single entry point + sub-command dispatch.
-
-## Sub-commands
-
-| Sub-command | Routes to | Purpose |
-|---|---|---|
-| `/judge solo` | `commands/judge/solo.md` | Standalone verdict on an existing diff — no implementer, no revision loop |
-| `/judge on-diff` | `commands/judge/on-diff.md` | Implementer→judge loop on a single change with a two-revision ceiling |
-| `/judge steps` | `commands/judge/steps.md` | Execute an ordered plan step by step, judge gate between steps |
-
-Sub-command names match the locked contract in
-[`docs/contracts/command-clusters.md`](../docs/contracts/command-clusters.md).
-The standalone reviewer surface lives at [`/review`](review-changes.md).
-
-## Dispatch
-
-1. Parse the user's argument: `/judge <sub-command> [args]`.
-2. Look up the sub-command in the table above.
-3. Load the body of the routed file and follow its `## Instructions`
-   section verbatim with the remaining args.
-4. If the sub-command is unknown or missing, print the table above and
-   ask:
-
-   > 1. solo — verdict only, no loop
-   > 2. on-diff — implementer→judge revision loop
-   > 3. steps — judge gate between ordered steps
-
-## Rules
-
-- **Do NOT commit, push, or open a PR** unless the sub-command
-  explicitly authorizes it.
-- **Do NOT chain sub-commands.** One `/judge <sub>` per turn.
-- If the user invokes `/judge` with no argument, **show the menu** —
-  do not guess which sub-command they meant.
-
-## See also
-
-- [`subagent-orchestration`](../skills/subagent-orchestration/SKILL.md)
-- [`/review`](review-changes.md) — human-oriented self-review (Reviewer-mode contract)
-- [`role-contracts`](../docs/guidelines/agent-infra/role-contracts.md#reviewer)

package/.agent-src/commands/knowledge/cross-repo.md

@@ -1,71 +0,0 @@
----
-model_tier: medium
-name: knowledge:cross-repo
-tier: 2
-cluster: knowledge
-sub: cross-repo
-description: Targeted, read-only retrieval over opted-in linked-project siblings (ADR-032 Option A). Pulls a shared type / API contract / config without bulk-including sibling files.
-skills: [file-editor]
-suggestion:
-  eligible: true
-  trigger_description: "what does the frontend expect, find the shared type in the other repo, check the sibling repo's API contract, /knowledge:cross-repo <query>"
-  trigger_context: "user needs context that lives in an attached sibling repo without copying its files in"
-workspaces:
-  - agent-config-maintainer
-packs:
-  - meta
----
-
-# /knowledge cross-repo
-
-Targeted, **read-only** retrieval across the IDE-attached sibling repos the user
-has opted into (`linked_projects[].include: true`). Returns a bounded set of
-matches — a shared type, an API contract the frontend consumes, a config the
-sibling owns — **without bulk-including** any sibling file. Implements
-[`cross-repo-retrieval`](../../../docs/contracts/cross-repo-retrieval.md) and
-stays inside [ADR-032](../../../docs/decisions/ADR-032-linked-projects-scope.md)
-Option A.
-
-## Prerequisites
-
-- Python 3.10+ on the host.
-- At least one sibling opted in (`agent-config linked-projects:list` shows them).
-  No opted-in sibling → the command is inert with a clear message.
-
-## Steps
-
-### 1. Parse the query
-
-`/knowledge cross-repo "<query>" [--path-scope <glob>]`. The query is one
-concept (≥ 1 term > 2 chars). A `--path-scope` glob narrows the search and is
-**required** for `large`-flagged siblings.
-
-### 2. Run the retrieval
-
-```bash
-python3 scripts/cross_repo_retrieve.py "<query>" [--path-scope <glob>] [--max-chunks 8]
-```
-
-The script reads opted-in siblings only, runs a targeted path-glob + content
-grep (never a full walk), redacts secrets/PII from every chunk, and returns the
-retrieval envelope.
-
-### 3. Present matches
-
-Render the table: `source_repo · path · freshness · why`. Each row names the
-source sibling, the path inside it, the last-commit/mtime freshness stamp, and
-why it matched. Use the chunks as *context*, never as files to copy wholesale.
-
-### 4. Honour the scope guards
-
-- A `large` sibling without `--path-scope` is skipped with a note — re-run with
-  a scope. Do not remove the guard.
-- A sibling not `include: true` is never read.
-
-## Rules
-
-- **Read-only.** Never write to a sibling. Out-of-root writes still pass the
-  host permission gate; this surface writes nothing.
-- **Opt-in only, targeted only.** No full-tree sweep; no implicit inclusion.
-- **Secrets never cross repos** — the chunk redactor runs before any text is shown.
-- **One concept per invocation.** Do not chain.

package/.agent-src/commands/knowledge/forget.md

@@ -1,96 +0,0 @@
----
-model_tier: medium
-name: knowledge:forget
-tier: 2
-cluster: knowledge
-sub: forget
-description: Drop a knowledge ingest from `agents/memory/knowledge/` by id prefix. Atomic, no partial state. Pinning protects from LRU eviction, not from explicit forget — pinned ingests are dropped the same.
-skills: [file-editor]
-suggestion:
-  eligible: true
-  trigger_description: "forget this ingest, drop the acme knowledge, remove the customer folder from agent memory"
-  trigger_context: "user wants to purge a specific knowledge ingest"
-workspaces:
-  - agent-config-maintainer
-packs:
-  - meta
----
-
-# /knowledge forget
-
-Removes one ingest from `agents/memory/knowledge/`. The id prefix
-must resolve to exactly one ingest — otherwise the command refuses
-and asks for a longer prefix.
-
-## Prerequisites
-
-- An ingest matching the prefix exists. Run `/knowledge list` first
-  to find it.
-- The id prefix is **8 characters or more** in practice — uuid7 ids
-  are unique on the first 8 hex chars within a single namespace,
-  and the dispatcher rejects ambiguous prefixes.
-
-## Steps
-
-### 1. Parse the argument
-
-```
-/knowledge forget <ingest-id-prefix>
-```
-
-If the prefix is missing → print the usage line above and stop. Do
-**not** guess — there is no safe default.
-
-### 2. Run the implementation
-
-```bash
-python3 packages/core/installer/python/knowledge_ingest.py \
-    forget <ingest-id-prefix>
-```
-
-The module:
-
-- Resolves the prefix; rejects if zero or multiple matches.
-- Removes the entire `agents/memory/knowledge/<ingest-id>/` directory
-  in one `shutil.rmtree` — chunks, manifest, everything.
-- Returns the full ingest-id that was removed.
-
-### 3. Surface the result
-
-```
-> ✅ forgot 01a2b3c4-5d6e-7890-abcd-ef1234567890
->   source: /Users/maintainer/clients/acme
->   documents: 12, chunks: 47, bytes recovered: 318012
-```
-
-If the prefix matched zero ingests, surface a structured error:
-
-```
-> ❌ no ingest matches prefix: <prefix>
->    run /knowledge list to see what's there
-```
-
-If the prefix matched multiple ingests:
-
-```
-> ❌ ambiguous prefix: <prefix> — matches <n> ingests
->    use a longer prefix (first 8 hex chars is usually enough)
-```
-
-## Safety
-
-- **Destructive — no undo.** The chunked content is gone after this
-  command returns. The Python module does not move-to-trash; it
-  removes the directory.
-- The command **never** removes the parent `agents/memory/knowledge/`
-  directory itself — only the matching `<ingest-id>/` subdirectory.
-- The command **never** writes outside `agents/memory/knowledge/`.
-- Pinning is irrelevant here — pinning protects from LRU eviction,
-  not from explicit user-driven forget.
-
-## See also
-
-- [`/knowledge ingest`](ingest.md) — add a new ingest.
-- [`/knowledge list`](list.md) — find an id prefix.
-- [`local-knowledge-ingestion`](../../../docs/contracts/local-knowledge-ingestion.md)
-  § Storage — directory layout.

package/.agent-src/commands/knowledge/ingest.md

@@ -1,114 +0,0 @@
----
-model_tier: medium
-name: knowledge:ingest
-tier: 2
-cluster: knowledge
-sub: ingest
-description: Walk a local path (folder, .zip, single file), redact PII + secrets, chunk to 2 KB markdown, and persist into the agent memory namespace under `knowledge/<ingest-id>/`.
-skills: [file-editor]
-suggestion:
-  eligible: true
-  trigger_description: "ingest this folder, add these PDFs to the agent, point the agent at my customer files, /knowledge:ingest <path>"
-  trigger_context: "user wants the agent to know about a local folder or document without OAuth connectors"
-workspaces:
-  - agent-config-maintainer
-packs:
-  - meta
----
-
-# /knowledge ingest
-
-Indexes a local path into the agent's memory namespace. Local-only,
-single-user, bounded. Implements the
-[`local-knowledge-ingestion`](../../../docs/contracts/local-knowledge-ingestion.md)
-contract.
-
-## Prerequisites
-
-- Python 3.10+ available on the host.
-- Path is local — absolute or relative to the project root. Remote
-  schemes are rejected.
-- `markitdown` is on `PATH` (or `--markitdown=<bin>` supplied) **if**
-  the corpus contains binary formats (PDF, DOCX, XLSX, EPUB, PPTX,
-  images). Pure markdown / text corpora work without it.
-
-## Steps
-
-### 1. Parse the argument
-
-The user invokes `/knowledge ingest <path> [--no-redact]
-[--markitdown=<bin>]`. The path is the first positional argument.
-
-If `<path>` is missing → print the usage line above and stop.
-
-### 2. Pre-flight checks
-
-Run before any file is touched:
-
-- Reject `http://`, `https://`, `s3://`, `gs://`, `azure://`.
-- Resolve the path; reject if it does not exist.
-- If the path is a `.zip`, the implementation unpacks into `$TMPDIR`
-  and cleans up after the call returns — no manual handling needed.
-
-### 3. Run the implementation
-
-```bash
-python3 packages/core/installer/python/knowledge_ingest.py \
-    ingest <path> [--no-redact] [--markitdown=<bin>]
-```
-
-The module enforces:
-
-- ≤ 1000 documents per call
-- ≤ 20 MB per file
-- ≤ 500 MB total namespace footprint (LRU eviction by `last_touched`)
-- ≤ 10 directories deep
-
-Crossing any bound is a **hard reject** with a structured error
-naming the bound and the observed value. No partial writes.
-
-### 4. Surface the result
-
-The module prints a JSON manifest. Summarize in chat:
-
-```
-> ✅ ingested <ingest-id-short> from <source>
->   documents: <n>, chunks: <m>, bytes_stored: <b>
->   PII redacted: EMAIL=<x>, PHONE=<y>, IBAN=<z>, CC=<a>, SSN=<b>
->   secrets redacted: <s>
->   skipped: <count> unsupported MIME
-```
-
-If `redacted: false` (user passed `--no-redact`), surface a
-**non-omittable warning** that the corpus may contain PII or secrets
-in plaintext.
-
-### 5. Hand off
-
-Show the user how to retrieve:
-
-```
-> Next steps:
-> 1. /knowledge list             — see all ingests + sizes
-> 2. /knowledge list --pin <id>  — pin so it survives LRU eviction
-> 3. /knowledge forget <id>      — drop this ingest atomically
-```
-
-## Safety
-
-- The command **never** fetches the network. Remote-URL rejection is
-  the first input check, before any path resolution.
-- The command **never** writes outside `agents/memory/knowledge/`.
-- Redaction defaults to ON. `--no-redact` is a deliberate maintainer
-  opt-out, not a UX default.
-- LRU eviction is by **ingest**, never by chunk — partial ingests
-  cannot exist in the namespace.
-
-## See also
-
-- [`local-knowledge-ingestion`](../../../docs/contracts/local-knowledge-ingestion.md)
-  — contract (input shapes, bounds, storage, redaction).
-- [`/knowledge list`](list.md) — what's been ingested, pinning.
-- [`/knowledge forget`](forget.md) — drop an ingest atomically.
-- [`markitdown`](../../skills/markitdown/SKILL.md) — peer-side adapter
-  for PDF / DOCX / XLSX / images.

package/.agent-src/commands/knowledge/list.md

@@ -1,94 +0,0 @@
----
-model_tier: medium
-name: knowledge:list
-tier: 2
-cluster: knowledge
-sub: list
-description: List existing knowledge ingests in `agents/memory/knowledge/` (table or JSON); pin / unpin by id prefix to control LRU eviction.
-skills: [file-editor]
-suggestion:
-  eligible: true
-  trigger_description: "what knowledge has the agent ingested, list ingested folders, show knowledge memory, pin this ingest"
-  trigger_context: "user wants to inspect or curate the local knowledge namespace"
-workspaces:
-  - agent-config-maintainer
-packs:
-  - meta
----
-
-# /knowledge list
-
-Shows everything currently held under `agents/memory/knowledge/`,
-sorted by `created_at` ascending. Read-only by default; `--pin` and
-`--unpin` mutate the `pinned` flag on a single ingest by id prefix.
-
-## Prerequisites
-
-- Python 3.10+ on the host.
-- `agents/memory/knowledge/` exists or is empty — the command treats
-  "empty" as a normal result, not an error.
-
-## Steps
-
-### 1. Parse the argument
-
-```
-/knowledge list [--format=table|json] [--pin <id-prefix>]
-                [--unpin <id-prefix>]
-```
-
-- `--format` defaults to `table` (ASCII; human-readable).
-- `--pin` and `--unpin` are mutually exclusive in spirit; the CLI
-  accepts only one per call.
-
-### 2. Run the implementation
-
-```bash
-python3 packages/core/installer/python/knowledge_ingest.py \
-    list [--format=...] [--pin <id-prefix>] [--unpin <id-prefix>]
-```
-
-### 3. Surface the result
-
-**Table view** columns:
-
-```
-ID        DOCS  CHUNKS  BYTES   PINNED  REDACTED  CREATED              SOURCE
-01a2b3c4  12    47      318012  no      yes       2026-05-25T08:14:02  /Users/maintainer/clients/acme
-```
-
-**JSON view** — full manifest array, one object per ingest. Useful for
-piping into `jq` or `/memory load` flows.
-
-**Pin / unpin** — one-line confirmation:
-
-```
-> ✅ pinned 01a2b3c4
-> ✅ unpinned 01a2b3c4
-```
-
-Pinning protects an ingest from LRU eviction when the 500 MB
-namespace cap is crossed.
-
-### 4. Empty namespace
-
-Print `(no ingests)` and offer the next step:
-
-```
-> 1. /knowledge ingest <path>  — point at a local folder, .zip, or file
-```
-
-## Safety
-
-- Read-only by default; only `--pin` / `--unpin` mutate state, and
-  both edit a single flag on a single manifest file.
-- The id prefix must be **unambiguous** — if it matches more than one
-  ingest, the command rejects with a structured error and asks for a
-  longer prefix.
-
-## See also
-
-- [`/knowledge ingest`](ingest.md) — add a new ingest.
-- [`/knowledge forget`](forget.md) — drop an ingest atomically.
-- [`local-knowledge-ingestion`](../../../docs/contracts/local-knowledge-ingestion.md)
-  § Bounds — when LRU fires.

package/.agent-src/commands/knowledge.md

@@ -1,69 +0,0 @@
----
-model_tier: medium
-name: knowledge
-tier: 2
-description: Knowledge orchestrator — routes to ingest, list, forget. Local-only file ingestion into the agent memory namespace.
-cluster: knowledge
-type: orchestrator
-suggestion:
-  eligible: true
-  trigger_description: "ingest local folder, add my customer files, point the agent at this folder, list ingested knowledge, forget that ingest"
-  trigger_context: "user wants to feed local files (folder, .zip, single doc) to the agent without OAuth connectors"
-workspaces:
-  - agent-config-maintainer
-packs:
-  - meta
----
-
-# /knowledge
-
-Top-level orchestrator for the `/knowledge` family — the **local
-knowledge ingestion** cluster. Single-user, single-machine, no OAuth,
-no remote fetches. Feeds local files (folder · `.zip` · single PDF /
-DOCX / XLSX / MD / TXT / image) into the existing memory namespace
-under the `knowledge/` prefix.
-
-Anchors: [`local-knowledge-ingestion`](../docs/contracts/local-knowledge-ingestion.md)
-contract — input shapes, bounds, redaction defaults, storage layout.
-
-## Sub-commands
-
-| Sub-command | Routes to | Purpose |
-|---|---|---|
-| `/knowledge ingest` | `commands/knowledge/ingest.md` | Walk a local path, redact, chunk, persist to `agents/memory/knowledge/<ingest-id>/` |
-| `/knowledge list` | `commands/knowledge/list.md` | List existing ingests (table or JSON); pin / unpin by id prefix |
-| `/knowledge forget` | `commands/knowledge/forget.md` | Drop a single ingest by id prefix (atomic, no partial state) |
-| `/knowledge cross-repo` | `commands/knowledge/cross-repo.md` | Targeted read-only retrieval over opted-in linked-project siblings (ADR-032 Option A) |
-
-Sub-command names match the locked contract in
-[`docs/contracts/command-clusters.md`](../docs/contracts/command-clusters.md).
-
-## Dispatch
-
-1. Parse the user's argument: `/knowledge <sub-command> [args]`.
-2. Look up the sub-command in the table above.
-3. Load the body of the routed file and follow its `## Steps` section
-   verbatim with the remaining args.
-4. If the sub-command is unknown or missing, print the table above and ask:
-
-   > 1. ingest — point at a local folder, `.zip`, or file
-   > 2. list — show what's already ingested (`--pin` / `--unpin` to flag)
-   > 3. forget — drop an ingest by id prefix
-   > 4. cross-repo — targeted read-only retrieval over opted-in siblings
-
-## Rules
-
-- **Local-only.** Remote URLs (`http://`, `https://`, `s3://`, `gs://`,
-  `azure://`) are rejected at the input validator. Never fetch the
-  network from this cluster.
-- **Bounds are non-negotiable** — see the contract. ≤ 1000 docs, ≤ 20 MB
-  per file, ≤ 500 MB namespace (LRU eviction at the namespace level).
-- **Redaction defaults to ON.** Five PII classes (email, phone, IBAN,
-  credit card, SSN-shape) and five secret patterns (AWS, GitHub PAT,
-  OpenAI key, private-key blocks, generic high-entropy `api_key=...`
-  assignments) are replaced before storage. `--no-redact` exists for
-  trusted maintainer corpora; it never becomes the default.
-- **Do NOT commit, push, or open a PR** unless the user explicitly asks.
-- **Do NOT chain sub-commands.** One `/knowledge <sub>` per turn.
-- If the user invokes `/knowledge` with no argument, **show the menu** —
-  do not guess which sub-command they meant.