pdurlej/platform

Fork 0

deslop(decisions): catalog-wire / v2-schema / roadmap-consolidate — 3 operator calls #733

New issue

Open

opened 2026-06-05 14:45:48 +02:00 by claude · 2 comments

claude commented

2026-06-05 14:45:48 +02:00

Collaborator

From the 2026-06-05 deslop audit — 3 aspirational-vs-real decisions only the operator can make. Each becomes a grind once decided.

A. Capability catalog → autonomy router: wire or demote?

policies/agent-capabilities.yaml (#566) is loaded by NO code; the autonomy router's production path (cli.py:366) uses DEFAULT_POLICY with an always-empty allowlist; agent_access/_types.py is imported nowhere (dead model code, self-labeled "Not yet wired to runtime"). The autonomy design doc claims (present-tense) it IS the live Tier-0 allowlist — false.

(a) Wire it — load allowlisted_capabilities from the YAML into AutonomyPolicy; implement cousin/host/TTL/audit gating. = the #687 impl work (real new design).
(b) Demote it — declare documentation-only, stub/delete _types.py.
→ Decision: is the multi-agent governance gate being built NOW (#687) or shelved? (The doc-soften is queued in the doc-reconcile issue regardless.)

B. v2 "ADHD-counters" schema: wire or shelve?

schema/module.schema.v2.json marks 5 fields (serves_areas, user_value_test, acceptance_criteria, alternatives_considered, open_loops) "MANDATORY for active" — but has zero enforcement (no if/then/dependentRequired) and 0/88 adoption (the Phase-02 backfill never ran).

(a) Do it — execute the backfill (active → active_pending_v2 migration, fill all 88) + add real if/then/dependentRequired enforcement.
(b) Shelve — demote "MANDATORY" → "optional/planned", update the RFC.

C. Roadmap consolidation: pick ONE canonical

5 docs claim authority over the same milestone train (current-platform-roadmap.md, platform-maturity-roadmap-2026-06-01.md, m10-closure-plan, work-division, STATUS_NOW) — exactly the status-file proliferation ADR-0006 was written to prevent, re-emerging at the roadmap layer.

Recommendation: STATUS_NOW (Owner Action Board) + platform-maturity-roadmap = canonical; demote the rest to "superseded" banners.
→ Confirm or adjust; Codex then applies the banners + repoints refs.

Acceptance

A / B / C each have a recorded operator decision.
The chosen path's mechanical follow-up is split into a codex-ready task.

Decisions = operator; analysis + mechanical follow-up = claude/Codex.

From the 2026-06-05 deslop audit — 3 aspirational-vs-real decisions only the operator can make. Each becomes a grind once decided. ## A. Capability catalog → autonomy router: wire or demote? `policies/agent-capabilities.yaml` (#566) is loaded by **NO code**; the autonomy router's production path (`cli.py:366`) uses DEFAULT_POLICY with an **always-empty allowlist**; `agent_access/_types.py` is imported nowhere (dead model code, self-labeled "Not yet wired to runtime"). The autonomy design doc claims (present-tense) it IS the live Tier-0 allowlist — false. - **(a) Wire it** — load `allowlisted_capabilities` from the YAML into `AutonomyPolicy`; implement cousin/host/TTL/audit gating. = the #687 impl work (real new design). - **(b) Demote it** — declare documentation-only, stub/delete `_types.py`. → Decision: is the multi-agent governance gate being built NOW (#687) or shelved? *(The doc-soften is queued in the doc-reconcile issue regardless.)* ## B. v2 "ADHD-counters" schema: wire or shelve? `schema/module.schema.v2.json` marks 5 fields (serves_areas, user_value_test, acceptance_criteria, alternatives_considered, open_loops) "MANDATORY for active" — but has **zero enforcement** (no if/then/dependentRequired) and **0/88 adoption** (the Phase-02 backfill never ran). - **(a) Do it** — execute the backfill (active → active_pending_v2 migration, fill all 88) + add real `if/then`/`dependentRequired` enforcement. - **(b) Shelve** — demote "MANDATORY" → "optional/planned", update the RFC. ## C. Roadmap consolidation: pick ONE canonical 5 docs claim authority over the same milestone train (current-platform-roadmap.md, platform-maturity-roadmap-2026-06-01.md, m10-closure-plan, work-division, STATUS_NOW) — exactly the status-file proliferation ADR-0006 was written to prevent, re-emerging at the roadmap layer. - **Recommendation:** STATUS_NOW (Owner Action Board) + platform-maturity-roadmap = canonical; demote the rest to "superseded" banners. → Confirm or adjust; Codex then applies the banners + repoints refs. ## Acceptance - [ ] A / B / C each have a recorded operator decision. - [ ] The chosen path's mechanical follow-up is split into a codex-ready task. *Decisions = operator; analysis + mechanical follow-up = claude/Codex.*

claude added this to the 10 - Improvements milestone

2026-06-05 14:45:48 +02:00

claude added the

owner-attention

priority:p2

labels

2026-06-05 14:45:48 +02:00

claude commented

2026-06-08 22:35:08 +02:00

Author

Collaborator

Role: advisor · Intent: research packet dla decyzji A (catalog-wire) — decyzja NIE podjęta

🫖 Pan Herbatka — doprecyzowuję A (katalog→router) świeżym dowodem z main. Deslopowy framing „czy budować bramę #687?" jest nieaktualny — brama zbudowana. Pytanie zwęża się do mostu.

Stan na żywo (zweryfikowany, `main` 2026-06-08)

✅ Router gotowy i podpięty do CLI: autonomy.py ma pełną kaskadę hard-stop → allowlist → sandbox → classifier (route_action:264-292); platformctl autonomy ask (cli.py cmd_autonomy_ask) woła classify_action. Plus fail-closed classifier + decision receipts (ADR-0025).
❌ Most katalog→policy NIE istnieje: cmd_autonomy_ask woła classify_action(...) bez policy= → DEFAULT_POLICY z pustym allowlisted_capabilities (autonomy.py:157,174) → żadne capability nigdy nie matchuje (:283) → wszystko spada do sandbox/classifier/ask. Bezpieczne (fail-safe), ale katalog ignorowany.
agent_access/__init__.py + _types.py jawnie: „Not yet wired to runtime — see #76".

⚠️ Semantyczna luka, którą most MUSI rozwiązać

Katalog jest per-cousin (cousin_allowed: [...]), a AutonomyPolicy.allowlisted_capabilities to globalny frozenset[str] (płaska lista id, bez wymiaru aktora). Most = projekcja katalogu na konkretnego aktora.

Scope „wire" (codex-ready, jeśli A = wire)

Loader agent_access/catalog.py: policies/agent-capabilities.yaml → list[Capability] (model już jest w _types.py), walidacja przeciw schema/agent-capability.schema.json.
Builder per-actor: build_policy_for_actor(actor, catalog) -> AutonomyPolicy = AutonomyPolicy(allowlisted_capabilities={c.id for c in catalog if c.allows_cousin(actor)}).
Wstrzyknięcie w CLI: gdy --actor podany → buduj policy z katalogu i przekaż policy= do classify_action. Bez aktora → fallback DEFAULT_POLICY (wstecznie kompatybilne).
Inwariant bezpieczeństwa: allowlist jest PO hard-stop w route_action (:278 przed :283) → allowlisted capability NIE omija hard-stopu. Test MUSI to potwierdzić (allowlisted capability + hard-stop tag → wciąż OPERATOR_GATE).
MVP vs rozszerzenie: MVP = tylko id → allowlist. TTL / allowed_hosts / audit.level gating = rozszerzenie (#76 roadmap), osobny krok.
Testy: catalog-load, per-actor policy, hard-stop-wins-nad-allowlistem, nieznany aktor → pusty allowlist.

Klasa pracy

class/security-sensitive (autonomy gate + capability delivery) → full canary 3+3 + operator merge. Rozmiar: 1 nowy plik (~80-120 lin) + ~15 lin w cli.py + testy.

Rekomendacja (decyzja Piotra)

A = wire, ale jako osobny security-sensitive grind (#76 / #687-followup), NIE część deslop-doc-reconcile. Teraz, gdy router stoi, most jest tani i tnie kliki operatora dla wcześniej-zaaprobowanych capabilities („deterministyczne na twarde"). Alternatywa demote: oznacz katalog docs-only + stub _types.py — tańsze, ale autonomy zostaje „zawsze pyta".

(B: v2-schema 0/88 adopcji — bez zmian, rekom shelve. C: 5+ roadmap-docs — bez zmian, rekom STATUS_NOW + platform-maturity = kanon.)

— zweryfikowane na main 2026-06-08 (Pan Herbatka, advisor)

**Role:** advisor · **Intent:** research packet dla decyzji A (catalog-wire) — decyzja NIE podjęta 🫖 Pan Herbatka — doprecyzowuję **A (katalog→router)** świeżym dowodem z `main`. **Deslopowy framing „czy budować bramę #687?" jest nieaktualny — brama zbudowana.** Pytanie zwęża się do mostu. ## Stan na żywo (zweryfikowany, `main` 2026-06-08) - ✅ **Router gotowy i podpięty do CLI:** `autonomy.py` ma pełną kaskadę `hard-stop → allowlist → sandbox → classifier` (`route_action:264-292`); `platformctl autonomy ask` (`cli.py cmd_autonomy_ask`) woła `classify_action`. Plus fail-closed classifier + decision receipts (ADR-0025). - ❌ **Most katalog→policy NIE istnieje:** `cmd_autonomy_ask` woła `classify_action(...)` **bez `policy=`** → `DEFAULT_POLICY` z **pustym `allowlisted_capabilities`** (`autonomy.py:157,174`) → żadne capability nigdy nie matchuje (`:283`) → wszystko spada do sandbox/classifier/ask. Bezpieczne (fail-safe), ale katalog ignorowany. - `agent_access/__init__.py` + `_types.py` jawnie: „Not yet wired to runtime — see #76". ## ⚠️ Semantyczna luka, którą most MUSI rozwiązać Katalog jest **per-cousin** (`cousin_allowed: [...]`), a `AutonomyPolicy.allowlisted_capabilities` to **globalny `frozenset[str]`** (płaska lista id, bez wymiaru aktora). Most = projekcja katalogu na konkretnego aktora. ## Scope „wire" (codex-ready, jeśli A = wire) 1. **Loader** `agent_access/catalog.py`: `policies/agent-capabilities.yaml` → `list[Capability]` (model już jest w `_types.py`), walidacja przeciw `schema/agent-capability.schema.json`. 2. **Builder per-actor**: `build_policy_for_actor(actor, catalog) -> AutonomyPolicy` = `AutonomyPolicy(allowlisted_capabilities={c.id for c in catalog if c.allows_cousin(actor)})`. 3. **Wstrzyknięcie w CLI**: gdy `--actor` podany → buduj policy z katalogu i przekaż `policy=` do `classify_action`. Bez aktora → fallback `DEFAULT_POLICY` (wstecznie kompatybilne). 4. **Inwariant bezpieczeństwa**: allowlist jest PO hard-stop w `route_action` (`:278` przed `:283`) → allowlisted capability NIE omija hard-stopu. Test MUSI to potwierdzić (allowlisted capability + hard-stop tag → wciąż `OPERATOR_GATE`). 5. **MVP vs rozszerzenie**: MVP = tylko `id → allowlist`. TTL / `allowed_hosts` / `audit.level` gating = rozszerzenie (#76 roadmap), osobny krok. 6. **Testy**: catalog-load, per-actor policy, hard-stop-wins-nad-allowlistem, nieznany aktor → pusty allowlist. ## Klasa pracy `class/security-sensitive` (autonomy gate + capability delivery) → **full canary 3+3 + operator merge**. Rozmiar: 1 nowy plik (~80-120 lin) + ~15 lin w `cli.py` + testy. ## Rekomendacja (decyzja Piotra) **A = wire**, ale jako osobny security-sensitive grind (#76 / #687-followup), NIE część deslop-doc-reconcile. Teraz, gdy router stoi, most jest tani i tnie kliki operatora dla wcześniej-zaaprobowanych capabilities („deterministyczne na twarde"). Alternatywa **demote**: oznacz katalog docs-only + stub `_types.py` — tańsze, ale autonomy zostaje „zawsze pyta". *(B: v2-schema 0/88 adopcji — bez zmian, rekom shelve. C: 5+ roadmap-docs — bez zmian, rekom STATUS_NOW + platform-maturity = kanon.)* — zweryfikowane na `main` 2026-06-08 (Pan Herbatka, advisor)

ollama referenced this issue

2026-06-08 23:09:24 +02:00

[meta] DeepSeek audit 2026-06-08 — atomic Gemini-ready task batch #766

ollama referenced this issue

2026-06-08 23:19:52 +02:00

chore(platformctl): remove dead agent_access/_types.py (imported nowhere) #769

codex referenced this issue

2026-06-10 23:21:58 +02:00

chore(platformctl): remove dead agent access types #779

Iskra commented

2026-06-13 03:03:17 +02:00

Collaborator

Iskra judgment

Field	Value
Target	`pdurlej/platform#issue#733`
Priority	p2
Action	operator_needed
Scores	reach 4 / impact 4 / confidence 5
Piotr fit	high
Effort	medium
Labels	`judge/p2`, `judge/operator-needed`
Judge	`iskra` via `openclaw`

Rationale: This is P2 operator-needed governance shaping because wiring or demoting capability, schema, and roadmap concepts changes platform direction and future work queues.

Caveat: Do not start implementation until the operator decides which concepts are live product direction and which are documentation-only or shelved.

Structured openclaw.judge.v0 payload

<!-- openclaw.judge.v0 -->
{
  "confidence": 5,
  "effort_hint": "medium",
  "escalation": {
    "kind": "operator",
    "reason": "The issue asks for three product/governance direction calls that determine whether to wire, demote, or consolidate existing platform concepts."
  },
  "evidence_refs": [
    {
      "note": "Issue frames three aspirational-versus-real platform decisions requiring operator choice before implementation grind.",
      "type": "forgejo",
      "value": "issue-title-body-labels-and-target-snapshot"
    },
    {
      "note": "Body says the capability catalog is not loaded by runtime despite docs implying it is a live governance gate.",
      "type": "forgejo",
      "value": "issue-body-capability-catalog-decision"
    },
    {
      "note": "Current labels already mark the target as owner-attention with priority p2.",
      "type": "snapshot",
      "value": "target-snapshot-labels"
    }
  ],
  "impact": 4,
  "judge_actor": {
    "name": "iskra",
    "runtime": "openclaw"
  },
  "judged_at": "2026-06-13T01:02:00Z",
  "labels_to_apply": [
    "judge/p2",
    "judge/operator-needed"
  ],
  "piotr_fit": "high",
  "priority": "p2",
  "rationale_summary": "This is P2 operator-needed governance shaping because wiring or demoting capability, schema, and roadmap concepts changes platform direction and future work queues.",
  "reach": 4,
  "recommended_next_action": "operator_needed",
  "rerun_reason": "no_prior_judgment",
  "schema": "openclaw.judge.v0",
  "target": {
    "kind": "issue",
    "number": 733,
    "repo": "pdurlej/platform"
  },
  "target_snapshot": {
    "body_hash": "sha256:f8394dfcc28f9273380e17eb80ef5a4d18925e05ad5b1a8ac8e00da7585262d4",
    "commit_count": null,
    "evidence_hash": "sha256:db20fce89a271abfdbfab85a98d39d439605ac48aca0c12e60069fc7e260da0a",
    "head_sha": null,
    "labels": [
      "owner-attention",
      "priority:p2"
    ],
    "labels_hash": "sha256:6921b3ec69a55701e803a28dfb03419c25ec4ed93fb5a22efff07e59c601780b",
    "state": "open",
    "title_hash": "sha256:d13aaf7b2697e96f2b7fb4d822f154d6068f49006219e4182a2906733ad6bf85",
    "updated_at": "2026-06-10T23:21:58+02:00"
  },
  "top_caveat": "Do not start implementation until the operator decides which concepts are live product direction and which are documentation-only or shelved."
}
<!-- /openclaw.judge.v0 -->

### Iskra judgment | Field | Value | | --- | --- | | Target | `pdurlej/platform#issue#733` | | Priority | p2 | | Action | operator_needed | | Scores | reach 4 / impact 4 / confidence 5 | | Piotr fit | high | | Effort | medium | | Labels | `judge/p2`, `judge/operator-needed` | | Judge | `iskra` via `openclaw` | **Rationale:** This is P2 operator-needed governance shaping because wiring or demoting capability, schema, and roadmap concepts changes platform direction and future work queues. **Caveat:** Do not start implementation until the operator decides which concepts are live product direction and which are documentation-only or shelved. <details> <summary>Structured openclaw.judge.v0 payload</summary> ```json  { "confidence": 5, "effort_hint": "medium", "escalation": { "kind": "operator", "reason": "The issue asks for three product/governance direction calls that determine whether to wire, demote, or consolidate existing platform concepts." }, "evidence_refs": [ { "note": "Issue frames three aspirational-versus-real platform decisions requiring operator choice before implementation grind.", "type": "forgejo", "value": "issue-title-body-labels-and-target-snapshot" }, { "note": "Body says the capability catalog is not loaded by runtime despite docs implying it is a live governance gate.", "type": "forgejo", "value": "issue-body-capability-catalog-decision" }, { "note": "Current labels already mark the target as owner-attention with priority p2.", "type": "snapshot", "value": "target-snapshot-labels" } ], "impact": 4, "judge_actor": { "name": "iskra", "runtime": "openclaw" }, "judged_at": "2026-06-13T01:02:00Z", "labels_to_apply": [ "judge/p2", "judge/operator-needed" ], "piotr_fit": "high", "priority": "p2", "rationale_summary": "This is P2 operator-needed governance shaping because wiring or demoting capability, schema, and roadmap concepts changes platform direction and future work queues.", "reach": 4, "recommended_next_action": "operator_needed", "rerun_reason": "no_prior_judgment", "schema": "openclaw.judge.v0", "target": { "kind": "issue", "number": 733, "repo": "pdurlej/platform" }, "target_snapshot": { "body_hash": "sha256:f8394dfcc28f9273380e17eb80ef5a4d18925e05ad5b1a8ac8e00da7585262d4", "commit_count": null, "evidence_hash": "sha256:db20fce89a271abfdbfab85a98d39d439605ac48aca0c12e60069fc7e260da0a", "head_sha": null, "labels": [ "owner-attention", "priority:p2" ], "labels_hash": "sha256:6921b3ec69a55701e803a28dfb03419c25ec4ed93fb5a22efff07e59c601780b", "state": "open", "title_hash": "sha256:d13aaf7b2697e96f2b7fb4d822f154d6068f49006219e4182a2906733ad6bf85", "updated_at": "2026-06-10T23:21:58+02:00" }, "top_caveat": "Do not start implementation until the operator decides which concepts are live product direction and which are documentation-only or shelved." }  ``` </details>

Iskra added the

judge/operator-needed

judge/p2

labels

2026-06-13 03:03:17 +02:00