feat(registry): Postman-style agent editor with lifecycle transitions

[toadkicker]

Summary

• Replace browse-only Registry UI with a Postman-style agent authoring editor (sidebar + tabbed editor + JSON-LD panel)
• Hide all schema.org vocabulary from UI surfaces via schema_phrase() — algorithmic conversion using heck (strips schema: prefix, Action suffix, converts CamelCase to Title Case)
• Derive ExecutionTarget (Remote/Local/SubAgent) from endpoint URL scheme at load time — zero TOML changes to 200+ catalog agents
• Wire lifecycle state machine (Draft → Published → Unpublished → Published) via Tauri IPC with persistent sentinel-based tracking in published_to

What changed

papillon-shared

• Add schema_phrase() for schema.org → plain-English conversion (6 tests)
• Add ExecutionTarget enum derived from endpoint URL scheme
• Add AgentLifecycle enum (Draft / Published / Unpublished) on AgentInfo

Papillon frontend (apps/papillon/frontend)

• New /registry route: RegistryPage with sidebar + editor layout
• AgentSidebar: agents grouped by lifecycle, colored state dots, "+ New" button
• AgentEditor: tabbed editor (Input / Returns / Disclosure / Endpoint / Settings), action selector showing plain-English verbs, lifecycle badge
• JsonLdPanel: live JSON-LD advertisement preview, lifecycle-contextual action button wired to Tauri IPC

Papillon backend (apps/papillon)

• sign_and_publish_local command: transitions Draft/Unpublished → Published
• unpublish_local command: transitions Published → Unpublished
• lifecycle_from_published_to() helper: single derivation point for lifecycle state (5 tests), using pap://local and pap://local:unpublished sentinels
• IPC errors surface via registry.error signal

Registry app (apps/registry)

• Apply schema_phrase() to capability, returns, and disclosure display strings

Test Plan

• cargo test -p papillon-shared — 335 passed
• cargo test -p papillon — all lifecycle tests pass (5 new tests for lifecycle_from_published_to)
• cargo check --manifest-path apps/papillon/frontend/Cargo.toml --features wasm — clean
• Open registry page, create an agent, verify no schema: strings visible
• Sign & Publish → agent moves to Published section in sidebar
• Unpublish → agent moves to Unpublished section
• Re-publish → agent moves back to Published section

🤖 Generated with Claude Code

[greptile-apps]

Greptile Summary

This PR replaces the browse-only Registry UI with a Postman-style agent editor (sidebar + tabbed editor + JSON-LD panel), adds schema_phrase() for schema.org → plain-English conversion, derives ExecutionTarget from endpoint URL scheme, and wires a Draft → Published → Unpublished lifecycle state machine via Tauri IPC with sentinel-based persistence.

• P1 — ExecutionTarget always None: ExecutionTarget::derive() is implemented and tested but never called in def_to_agent_info, list_local_agents, ad_to_info (registry.rs), or web_service.rs. The UI's "Execution Target" badge will show "None" for every agent.
• P1 — Silent no-op on "Sign & Publish" for Draft agents: on_action in JsonLdPanel returns early without any user feedback when agent_did is None or empty — which is always the case for new Draft agents, the primary target of that button.

Confidence Score: 3/5

Two P1 defects make the core publish flow and execution-target feature non-functional as shipped.

Two independent P1 issues: ExecutionTarget is always None despite derive() existing, and the Sign & Publish button silently no-ops for any agent lacking a pre-existing DID (every new Draft agent). Both affect primary flows in the PR test plan. Multiple P1s pull the score below the 4/5 ceiling.

apps/papillon/src/commands/agents.rs (ExecutionTarget::None hardcoded), apps/papillon/frontend/src/components/registry/json_ld_panel.rs (silent no-op), apps/papillon/src/commands/registry.rs and apps/papillon/frontend/src/service/web_service.rs (same ExecutionTarget::None pattern).

Important Files Changed

Filename	Overview
apps/papillon/src/commands/agents.rs	Adds sign_and_publish_local and unpublish_local Tauri commands plus lifecycle_from_published_to helper; ExecutionTarget is hardcoded to None instead of being derived from the endpoint in all conversion paths.
apps/papillon/frontend/src/components/registry/json_ld_panel.rs	New JSON-LD preview panel with lifecycle-contextual action button; on_action silently no-ops for Draft agents without a DID — the primary publish flow — with no user-visible error.
apps/papillon/frontend/src/components/registry/agent_editor.rs	New Postman-style tabbed editor; tab count badges are non-reactive (captured once at render) and will not update when switching agents in the sidebar.
crates/papillon-shared/src/types.rs	Adds ExecutionTarget enum with derive() factory and AgentLifecycle enum; both are well-typed with sensible defaults and clean serde attributes.
crates/papillon-shared/src/schema_phrase.rs	New schema_phrase() utility with 6 unit tests covering prefix strip, suffix strip, CamelCase conversion, and edge cases — clean and well-tested.
apps/papillon/frontend/src/components/registry/agent_sidebar.rs	New sidebar grouping agents by lifecycle with colored dots; uses agent name as selection key which is fragile but consistent with the rest of the state model.
apps/papillon/frontend/src/components/registry/peer_browser.rs	Renamed from browser.rs; inlines former AgentCard but still strips schema: with trim_start_matches rather than the new schema_phrase(), leaving Action suffix and CamelCase un-normalized.
apps/papillon/src/commands/registry.rs	Minimal change to populate new AgentInfo fields; execution_target also hardcoded to None here.
apps/registry/src/ui/pages/agents.rs	Applies schema_phrase() to capability, returns, and disclosure display strings — straightforward and correct.

Sequence Diagram

sequenceDiagram
    participant UI as JsonLdPanel (Frontend)
    participant Bridge as Tauri Bridge
    participant Cmd as commands/agents.rs
    participant DB as Database

    Note over UI: User clicks Sign & Publish (Draft/Unpublished)
    UI->>UI: Check agent_did — returns early if None/empty (silent no-op)
    UI->>Bridge: invoke(sign_and_publish_local, { agent_did })
    Bridge->>Cmd: sign_and_publish_local(agent_did)
    Cmd->>DB: load_all_agents()
    DB-->>Cmd: Vec DynamicAgentDef
    Cmd->>Cmd: retain: remove LOCAL_UNPUBLISHED_SENTINEL
    Cmd->>Cmd: push: LOCAL_REGISTRY_URL
    Cmd->>DB: update_agent(def)
    Cmd-->>Bridge: Ok(AgentInfo lifecycle Published)
    Bridge-->>UI: Ok(AgentInfo)
    UI->>Bridge: invoke(list_local_agents)
    Bridge->>Cmd: list_local_agents()
    Cmd-->>Bridge: Vec AgentInfo
    Bridge-->>UI: agents list
    UI->>UI: registry.agents.set(agents) sidebar re-renders

Loading

Comments Outside Diff (3)

1. apps/papillon/src/commands/agents.rs, line 908-911 (link)

   ExecutionTarget always None — derive() never called in backend

   def_to_agent_info (line 908) and list_local_agents (line 917–920) both hard-code execution_target: ExecutionTarget::None even though ExecutionTarget::derive() is implemented and tested. The UI's "Execution Target" badge in EndpointTab will therefore always render "None" regardless of the actual endpoint URL. The same pattern is repeated in apps/papillon/src/commands/registry.rs and apps/papillon/frontend/src/service/web_service.rs.

   Prompt To Fix With AI

   This is a comment left during a code review.
   Path: apps/papillon/src/commands/agents.rs
   Line: 908-911
   
   Comment:
   **`ExecutionTarget` always `None` — derive() never called in backend**
   
   `def_to_agent_info` (line 908) and `list_local_agents` (line 917–920) both hard-code `execution_target: ExecutionTarget::None` even though `ExecutionTarget::derive()` is implemented and tested. The UI's "Execution Target" badge in `EndpointTab` will therefore always render "None" regardless of the actual endpoint URL. The same pattern is repeated in `apps/papillon/src/commands/registry.rs` and `apps/papillon/frontend/src/service/web_service.rs`.
   
   
   
   How can I resolve this? If you propose a fix, please make it concise.

   

2. apps/papillon/frontend/src/components/registry/json_ld_panel.rs, line 638-643 (link)

   Silent no-op when agent_did is None for Draft agents

   on_action returns early without any feedback when agent_did is None or empty. New Draft agents — the primary use-case for "Sign & Publish" — won't have a DID yet, so clicking the button silently does nothing. There is no toast, error message, or registry.error signal set to tell the user why the action failed. This also means the "Sign & Publish" button appears active but is functionally broken for any agent without a pre-existing DID.

   Prompt To Fix With AI

   This is a comment left during a code review.
   Path: apps/papillon/frontend/src/components/registry/json_ld_panel.rs
   Line: 638-643
   
   Comment:
   **Silent no-op when `agent_did` is `None` for Draft agents**
   
   `on_action` returns early without any feedback when `agent_did` is `None` or empty. New Draft agents — the primary use-case for "Sign & Publish" — won't have a DID yet, so clicking the button silently does nothing. There is no toast, error message, or `registry.error` signal set to tell the user why the action failed. This also means the "Sign & Publish" button appears active but is functionally broken for any agent without a pre-existing DID.
   
   How can I resolve this? If you propose a fix, please make it concise.

   

3. apps/papillon/frontend/src/components/registry/peer_browser.rs, line 724-728 (link)

   Inlined card still uses raw trim_start_matches("schema:") instead of schema_phrase()

   The inlined agent card in PeerBrowser strips the schema: prefix manually (.trim_start_matches("schema:")) but never strips the Action suffix or converts CamelCase to Title Case — inconsistent with the rest of the PR's stated goal of hiding schema.org vocabulary via schema_phrase(). Consider replacing with:

   (and adding use papillon_shared::schema_phrase; at the top of the file.)

   Prompt To Fix With AI

   This is a comment left during a code review.
   Path: apps/papillon/frontend/src/components/registry/peer_browser.rs
   Line: 724-728
   
   Comment:
   **Inlined card still uses raw `trim_start_matches("schema:")` instead of `schema_phrase()`**
   
   The inlined agent card in `PeerBrowser` strips the `schema:` prefix manually (`.trim_start_matches("schema:")`) but never strips the `Action` suffix or converts CamelCase to Title Case — inconsistent with the rest of the PR's stated goal of hiding schema.org vocabulary via `schema_phrase()`. Consider replacing with:
   
   (and adding `use papillon_shared::schema_phrase;` at the top of the file.)
   
   How can I resolve this? If you propose a fix, please make it concise.

   Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

   

Prompt To Fix All With AI

Fix the following 4 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 4
apps/papillon/src/commands/agents.rs:908-911
**`ExecutionTarget` always `None` — derive() never called in backend**

`def_to_agent_info` (line 908) and `list_local_agents` (line 917–920) both hard-code `execution_target: ExecutionTarget::None` even though `ExecutionTarget::derive()` is implemented and tested. The UI's "Execution Target" badge in `EndpointTab` will therefore always render "None" regardless of the actual endpoint URL. The same pattern is repeated in `apps/papillon/src/commands/registry.rs` and `apps/papillon/frontend/src/service/web_service.rs`.

```suggestion
        execution_target: papillon_shared::ExecutionTarget::derive(def.endpoint.as_deref()),
        lifecycle: lifecycle_from_published_to(&def.published_to),
```

### Issue 2 of 4
apps/papillon/frontend/src/components/registry/json_ld_panel.rs:638-643
**Silent no-op when `agent_did` is `None` for Draft agents**

`on_action` returns early without any feedback when `agent_did` is `None` or empty. New Draft agents — the primary use-case for "Sign & Publish" — won't have a DID yet, so clicking the button silently does nothing. There is no toast, error message, or `registry.error` signal set to tell the user why the action failed. This also means the "Sign & Publish" button appears active but is functionally broken for any agent without a pre-existing DID.

### Issue 3 of 4
apps/papillon/frontend/src/components/registry/agent_editor.rs:250-254
**Tab count badges are non-reactive — captured once at render**

`input_count()`, `returns_count()`, and `disclosure_count()` are called immediately and passed as `Option<usize>` to `TabButton`. Since `TabButton` accepts a plain `usize` (not a signal), these values are snapshotted at the time the `AgentEditor` is first composed. Switching to a different agent in the sidebar will update the tab content but leave the count badges stale.

Consider passing the closures as signals or wrapping in `Signal::derive()` so they re-evaluate when the `agent` signal changes.

### Issue 4 of 4
apps/papillon/frontend/src/components/registry/peer_browser.rs:724-728
**Inlined card still uses raw `trim_start_matches("schema:")` instead of `schema_phrase()`**

The inlined agent card in `PeerBrowser` strips the `schema:` prefix manually (`.trim_start_matches("schema:")`) but never strips the `Action` suffix or converts CamelCase to Title Case — inconsistent with the rest of the PR's stated goal of hiding schema.org vocabulary via `schema_phrase()`. Consider replacing with:
```suggestion
                            .map(|c| papillon_shared::schema_phrase(c))
```
(and adding `use papillon_shared::schema_phrase;` at the top of the file.)

_{Reviews (1): Last reviewed commit: "The implementation is complete. 12 commi..." | Re-trigger Greptile}

[greptile-apps]

Tab count badges are non-reactive — captured once at render

input_count(), returns_count(), and disclosure_count() are called immediately and passed as Option<usize> to TabButton. Since TabButton accepts a plain usize (not a signal), these values are snapshotted at the time the AgentEditor is first composed. Switching to a different agent in the sidebar will update the tab content but leave the count badges stale.

Consider passing the closures as signals or wrapping in Signal::derive() so they re-evaluate when the agent signal changes.

Prompt To Fix With AI

This is a comment left during a code review.
Path: apps/papillon/frontend/src/components/registry/agent_editor.rs
Line: 250-254

Comment:
**Tab count badges are non-reactive — captured once at render**

`input_count()`, `returns_count()`, and `disclosure_count()` are called immediately and passed as `Option<usize>` to `TabButton`. Since `TabButton` accepts a plain `usize` (not a signal), these values are snapshotted at the time the `AgentEditor` is first composed. Switching to a different agent in the sidebar will update the tab content but leave the count badges stale.

Consider passing the closures as signals or wrapping in `Signal::derive()` so they re-evaluate when the `agent` signal changes.

How can I resolve this? If you propose a fix, please make it concise.

[github-actions]

Benchmark Regression Report

PAP Protocol Benchmark Regression Check
========================================
Baseline: .bench-baseline/baseline.json
Threshold: 55%

  ed25519_keypair_generation                19.6 µs  (baseline: 17.1 µs, +14.4%)  [ok]
  did_key_derivation                         1.5 µs  (baseline: 1.4 µs, +10.1%)  [ok]
  mandate_create_sign                       23.9 µs  (baseline: 22.0 µs, +8.5%)  [ok]
  mandate_chain_verify_depth3              145.4 µs  (baseline: 128.0 µs, +13.6%)  [ok]
  sd_jwt_issue_5claims                      27.8 µs  (baseline: 25.1 µs, +11.0%)  [ok]
  sd_jwt_verify_disclose_3of5               50.4 µs  (baseline: 44.2 µs, +14.1%)  [ok]
  session_open_full_lifecycle              108.4 µs  (baseline: 100.3 µs, +8.0%)  [ok]
  receipt_create_cosign                     48.7 µs  (baseline: 44.2 µs, +10.0%)  [ok]
  federation_announce_local                 62.2 µs  (baseline: 55.8 µs, +11.5%)  [ok]


P99 Tail-Latency Check
----------------------
Results: target/p99_results.json
Threshold: 50%

  session_open_full_lifecycle(p99)         126.5 µs  (baseline: 500.0 µs, -74.6%)  [ok]
  mandate_chain_verify_depth3(p99)         167.0 µs  (baseline: 480.0 µs, -65.1%)  [ok]
  receipt_create_cosign(p99)                56.4 µs  (baseline: 210.0 µs, -73.1%)  [ok]

All benchmarks within 55% of baseline.

_{Threshold: 10% regression vs baseline from main}