feat(cli): deliberate TUI claim/start and direct navigation

.commitrail/INDEX.md

@@ -11,7 +11,7 @@ for current product capability.
 | [WS-AGENT-001](initiatives/WS-AGENT-001/OVERVIEW.md) | Planned | Neutral document tools plus OpenAI Agents SDK and OmniCoreAgent original-guide qualifications are delivered through the same public API, durable worker, parser and MinIO path. Retain the exact external provider/account and selected-image AWS evidence, then prove the broader hosted Flow contribution journey before deployment claims. |
 | [WS-DB-002](initiatives/WS-DB-002/OVERVIEW.md) | Complete | Shared UUIDv7 record generation, native-UUID relationships and fresh v0.1 baseline; natural-owner retry custody and aligned CI/local setup |
 | [WS-MCP-002](initiatives/WS-MCP-002/OVERVIEW.md) | Planned | 108 tools through WS-MCP-002-12 cover access, guide setup, contributor workspace, contribution policies, checker/artifact operations and task execution/submissions; task operations/imports and assembled release proof remain. |
-| [WS-CLI-001](initiatives/WS-CLI-001/OVERVIEW.md) | Planned | Public Go CLI setup, contributor/manager workflows, administrative and operational commands, Finance policies and contribution exports are detailed in the overview; -43 proves the connected local unpaid automated-acceptance journey. -44 prepares four native release archives and a gated manual publisher; -47 hardens its administrator-bypass guard. First publication and public-asset audit remain operational work. -46 adds the separate system-reader project catalogue; -48 adds contributor TUI navigation and -49 adds findings pages and confirmed verified guide downloads. Evidence remains a manual CLI handoff; TUI product writes/management, public human-review controls and hosted proof are separate. Current revision context never substitutes for historical evidence. |
+| [WS-CLI-001](initiatives/WS-CLI-001/OVERVIEW.md) | Planned | Public Go CLI setup, contributor/manager workflows, administrative and operational commands, Finance policies and contribution exports are detailed in the overview; -43 proves the connected local unpaid automated-acceptance journey. -44 prepares four native release archives and a gated manual publisher; -47 hardens its administrator-bypass guard. First publication and public-asset audit remain operational work. -46 adds the separate system-reader project catalogue; -48/-49 add contributor navigation, findings and verified guide downloads; -50 establishes the reading workbench. -51 bounds confirmed Claim/Start and direct collection navigation. TUI ZIP preparation/submission, automatic crash-safe recovery, contribution evidence, management, public human-review controls and hosted proof remain separate. Current revision context never substitutes for historical evidence. |
 | [WS-ARCH-001](initiatives/WS-ARCH-001/OVERVIEW.md) | Planned | Public admission-backed initial/remediation Submission entry and supported false-policy guide activation are implemented over selected production delivery. New false/pass effects require LIVE REV; a known pre-effect non-live refusal has bounded retry, with exhausted delivery dead-lettered. The [bounded local public/worker drill](initiatives/WS-ARCH-001/planning/PLAN.md#first-complete-contributor-milestone) starts from public guide inputs with recorded inference; hosted operation and live human review remain. |
 | [WS-ART-001](initiatives/WS-ART-001/OVERVIEW.md) | Planned | Verified input/output custody, exact-cause replacement and public admission-backed ZIP/Submission entry are delivered. The bounded local public/worker drill passed; hosted operation and setup-fault handling remain separate. |
 | [WS-AUTH-001](initiatives/WS-AUTH-001/OVERVIEW.md) | Planned | Exact fixed-router custody, checker remediation and checker retry are delivered. The ART Operator backend prerequisite adds exact system-Operator artifact reads and independently guarded verification retry; WS-CLI-001-43 consumes all nine public ART operations, while hosted S3 proof remains separate. Public entry and supported false-guide activation use current exact AUTH/SQL custody; human decision authority remains later work. |

.commitrail/initiatives/WS-CLI-001/OVERVIEW.md

@@ -66,10 +66,40 @@ ordered locked-guide selection with explicit verified single-document download.
 [WS-CLI-001-50](WS-CLI-001-50.md) replaces their contributor presentation with
 visible collections/sections, readable instructions and truthful page/evidence
 states; it changes no public operation or lifecycle authority.
-Contribution evidence remains an inert CLI handoff. Product mutations, management
-screens and operation-specific recovery remain separate TUI boundaries; this
+Contribution evidence remains an inert CLI handoff. Claim/start is bounded by
+-51 below; other product mutations, management screens and their recovery remain
+separate TUI boundaries; this
 interface does not depend on first binary publication or system-wide discovery.
 
+## Connected contributor workbench
+
+The human's post-merge evaluation of -50 found that the read-only workbench was
+not a usable complete workflow and collection/detail navigation was difficult.
+Presentation proof is not completion of the contributor journey. Continue in
+the existing TUI and typed API client, not a second client or simulated demo.
+
+1. [WS-CLI-001-51](WS-CLI-001-51.md): direct collection navigation and deliberate
+   claim/start using existing public writes, exact confirmation, caller-owned
+   replay keys and uncertain-outcome recovery. This is the current bounded work.
+2. ZIP preparation: explicit local package selection, summary/attestation,
+   actual intake feedback and exact replay. Preparation is not Submission.
+3. Submission and continuation: deliberate immutable creation, current outcome,
+   checker remediation and only API-supported human revision responses. No
+   automatic acceptance, hidden mutation route or invented reviewer workflow.
+4. Contribution evidence: inspect the contributor's returned records and awards
+   inside the workbench, retaining the distinction from fulfillment.
+
+Only the first boundary is specified for implementation now. Each following
+boundary needs inspected API/replay contracts, its own bounded record and native
+public-API proof before starting. A durable automatic recovery design is needed
+before replacing caller-owned saved keys with generated keys; the current CLI
+contract must not become an unsafe in-memory TUI retry token.
+
+Risk is L1 for mutation, cancellation, disclosure and local artifact handling.
+Preserve backend authority, locked lineage, immutable evidence and atomicity.
+No new API is required for claim/start. Backend gaps discovered for subsequent
+steps must be named and independently implemented, never hidden by a UI mock.
+
 ## Current boundary
 
 [WS-CLI-001-26](WS-CLI-001-26.md) adds deliberate draft-guide summary editing
@@ -78,7 +108,7 @@ and separate guide-bound review/revision policy configuration. The commands
 the existing public writes with caller-owned replay keys and policy `If-Match`.
 The backend owns manager authorization for new writes, policy semantics, replay and activation;
 editing/configuring a guide does not approve or activate it. Other public API
-client gaps and the broader TUI beyond -49 remain separate work. Binary distribution's
+client gaps and the broader TUI beyond -51 remain separate work. Binary distribution's
 workflow is prepared by -44; a protected release run and public-asset audit
 remain operational work.
 

.commitrail/initiatives/WS-CLI-001/WS-CLI-001-51.md

@@ -0,0 +1,128 @@
+# WS-CLI-001-51 — Claim and start from the contributor workbench
+
+- Initiative: `WS-CLI-001`
+- Durable disposition: `Planned`
+- Intended merge outcome: Let a contributor deliberately claim and start exact tasks through the existing public API, with discoverable navigation and honest uncertain-outcome recovery.
+
+## Intent
+
+The merged -50 workbench is an observation interface, not a complete contributor
+journey. The human found navigation difficult and missing actions made the UI
+feel temporary. Main `f16dfe055` exposes working `api.Client.ClaimTask` and
+`StartTask` and explicit CLI commands, but TUI `ReadClient` and `Model.key` do not
+expose them. No backend prerequisite is needed. The existing real-API TUI drill
+proves reads and fresh revocation only; it must not be relabelled mutation proof.
+
+## Bounded change
+
+Allowed: `cli/internal/tui/`, TUI wiring in `cli/internal/command/tui.go` and its
+affected tests, focused existing CLI native HTTP/public-API tests and an opt-in
+POST capture hook in their existing socket fixture, CLI README,
+the single terminal design document, affected roadmap/initiative/index claims.
+Reuse the existing typed mutation client, sanitizer, response validation and
+backend authorization. No backend/client transport/CI/dependency changes.
+
+Not allowed: upload/Submission/revision/Review/Finance mutations, automatic
+chained actions, polling, credentials/history storage, generic operation journal,
+shell execution, compatibility paths, new auth rules or weakening tests.
+
+## Design and decisions
+
+- Add visible `a actions` on freshly inspected Ready task or Mine Overview.
+  Offer only Claim for observed ready work, Start for observed claimed Mine.
+  Observed status determines UI relevance, never server authority.
+  Gate both on exact freshly fetched `ContributorTaskDetail`, not row status;
+  Mine rows carry assignment status (`active`), not task status (`claimed`).
+- Require a caller-owned UUID replay key saved outside this session, matching
+  the explicit CLI contract. Explain this requirement before entry. Do not
+  generate a transient key or imply crash recovery exists. Reason is omitted
+  consistently in this slice (JSON `{}` through existing client).
+  The UI cannot verify external saving: require user input and an explicit
+  save-warning acknowledgment at confirmation, with no generated fallback.
+- Confirm exact action, actor, project, task, key and available locked terms;
+  Enter in key entry only advances to confirmation. Only a deliberate `y` at a
+  usable terminal dispatches one POST. No request on selection or resize.
+- Freeze the packet while sending. Navigation/refresh cannot replace it or
+  dispatch another action. Quit cancels the client request, not server effects.
+  Unknown, malformed success and transport cancellation never mean rollback.
+- Preserve unknown packet in the session, with explicit same-packet replay and
+  inert CLI recovery arguments. No automatically retried POST; no new key after
+  uncertainty. Session loss clears protected work; retained recovery references
+  must not expose previously cleared descriptions/terms or bearer.
+- Confirmed receipt and subsequent read are separate outcomes. Do not erase a
+  successful claim/start because a refresh fails; do not automatically start
+  after claim. Reauthorize through existing API on every write/replay.
+- Use a separate write outcome handler, not `failed()`'s read-page reset.
+  Bind returned task/project to the frozen packet and claim assignment contributor
+  to the bound actor; the existing typed decoder does not check the latter two
+  against the caller's selection. Any mismatch is unknown, never success.
+  Complete canonical 4xx is a known denial of that request; transport, cancellation
+  and malformed responses remain unknown. A denied replay cannot retroactively
+  prove an earlier unknown request failed. Keep unknown history until a validated
+  receipt resolves it. On 401/403/concealed 404, clear protected descriptions and
+  terms, retaining only minimal inert replay references; session failure disables
+  replay in this session. Pending writes consume navigation/refresh/action keys.
+  Quit cancels the request and exits, never claims server rollback; the caller's
+  externally saved key is the restart recovery contract, not an in-memory journal.
+- Add visible direct collection shortcuts `1 Ready`, `2 Mine`, `3 Submissions`
+  from ordinary list/detail views and reverse Tab navigation. Text entry,
+  confirmations, pending operations and overlays must consume their own keys.
+  Preserve current keys and exact-resource cancellation semantics.
+
+The caller-owned key is an explicit initial limitation, not the desired final
+consumer UX. Automatic crash-safe operation recovery needs a separately bounded
+private-storage design; do not hide it in this action slice.
+
+## Acceptance criteria
+
+- Native TUI can inspect Ready, confirm claim, see receipt, inspect Mine, then
+  separately confirm start. Exact assignment/locked lineage remains backend-owned.
+- Decline, resize below minimum, repeated keys, read movement and text entry
+  cannot cause a write. Different task/project/actor cannot receive a late result.
+- Lost response retains original action/key and only explicit replay reuses it.
+  Read failure does not manufacture rejection, rollback or confirmed success.
+- Real signed public API/PostgreSQL native drill proves claim/start, exact-key
+  replay, ownership and fresh revoked-authority rejection; preserve old GET-only
+  drill. HTTP tests prove exact wire/no automatic retry and corrupt receipt.
+  Use the separate ready task from `create_task`, not `_prepared_packet`'s already
+  started task. Prove claim replay before start, then start replay before revoke:
+  claim replay after progression may validly return state-conflict. Assert
+  caller/project substitution is unknown and that a successful write plus failed
+  read still shows the confirmed write receipt without restoring protected text.
+- Go build/test/vet/race, native HTTP tests, layout/hostile text/download guards,
+  links, wording and full hosted proof remain required. Add a discriminating
+  hidden-confirmation/duplicate-dispatch regression; do not weaken old assertions.
+
+## Risk and review routing
+
+L1: contributor mutation, uncertainty and untrusted terminal content. Plan review
+before implementation; focused security, product/operations, architecture/reuse,
+QA/test-delta, CI integrity (shared native socket fixture) and docs review on a
+frozen candidate. Human focus: actions are
+obvious, claim and start remain separate, exact confirmation and uncertain-result
+copy are understandable. No merge without specific human approval.
+
+## Evidence
+
+Focused model regressions exercise confirmation, frozen request identity,
+substituted receipts, denied replay, protected clearing, bounded paste and
+cross-resource navigation. Native HTTP tests exercise the built binary and
+exact POST bytes. The separate signed public-API drill exercises claim and
+start, same-key replay and live grant revocation against isolated PostgreSQL;
+the earlier GET-only drill remains unchanged. These are distinct proof
+boundaries, not evidence of the remaining submission journey. Exact-head
+execution and reviewer results belong in the PR trust summary.
+
+Independent review exposed a late-read/session-denial interleaving: opening
+replay while a refresh was pending could leave confirmation actionable after
+the read blocked the session. Dispatch must recheck session and replay
+eligibility, not rely solely on the earlier modal entry check. A focused
+regression must reject the old dispatch behavior. Same-collection shortcuts
+from detail must also return to the list rather than silently doing nothing.
+
+## Reconciliation and remaining journey
+
+This chunk closes Ready -> claimed -> in_progress inside the TUI. Subsequent
+bounded work: ZIP preparation with feedback and safe replay; immutable submission
+and outcome/revision navigation; contributor evidence and awards. Those are not
+delivered by this chunk. Human Review controls require their own public API proof.

cli/README.md

@@ -46,16 +46,35 @@ terms remain available separately. Workstream rechecks each public read.
 |---|---|
 | Arrows / Enter | Select a row or scroll detail; Enter inspects freshly |
 | Tab / Esc | Change section / go back or cancel |
+| Shift+Tab | Previous collection or detail section |
+| `1` / `2` / `3` | Go directly to Ready / Mine / Submissions from ordinary views |
+| `a` in task Overview | Open Claim or Start when the freshly inspected task supports that action |
 | `g` | Fetch the project picker again |
 | `n` / `r` | Fetch the next page / refresh the current view |
 | `/` | Filter loaded names/titles locally; Enter finishes and Esc cancels |
 | `c` / `?` | Show inert CLI arguments / keyboard help |
 | `d` in Guide | Enter an existing download directory; Enter reviews the scrollable exact target, `y` confirms |
-| `q` / Ctrl+C | Quit and cancel outstanding client reads; `q` is text during text entry, Ctrl+C always quits |
-
-The workspace makes no product mutation requests: it does not claim/start work,
-prepare or submit ZIPs, export records, poll, acquire tokens or save local
-history. Guide metadata is not proof of downloaded-byte integrity. In Guide,
+| `q` / Ctrl+C | Quit and cancel outstanding client requests, not committed server effects; `q` is text during text entry, Ctrl+C always quits |
+
+Claim and Start are separate deliberate actions through the existing public
+API. Inspect an exact Ready task to claim it; inspect a claimed Mine task to
+start it. Supply a UUID key you have saved outside this session, review the
+exact action/actor/project/task/key, press `s` to acknowledge the save warning,
+then `y` to confirm. Enter alone never sends the action. The workbench does not generate or persist these keys. Save the
+action, task ID and key together for recovery; this slice omits the optional
+reason. A read or displayed action is not write authority.
+
+There is no automatic retry or claim-to-start chain. A lost response means
+unknown outcome, not rollback. Explicit replay uses the unchanged action, task
+and key (`p`, then review and acknowledge with `s` and confirm with `y`);
+a later denial cannot prove the earlier request failed. Quit cancels
+the client request, not any committed server effect. After restarting, use the
+explicit CLI with your saved inputs and the same API origin and actor. Do not
+substitute a fresh key or infer success from a refreshed list. Automatic
+crash-safe recovery is not implemented.
+
+The workspace does not prepare or submit ZIPs, export records, poll, acquire
+tokens or save local history. Guide metadata is not proof of downloaded-byte integrity. In Guide,
 select one document, enter an existing real directory and confirm its label,
 document/task IDs, path, UUID filename, byte count and hash. Verified bytes are
 published under that filename without overwriting an existing file. Cancelled
@@ -72,11 +91,12 @@ automatically.
 A denial clears affected content; expired authentication clears the session's
 work data. A failed current read clears its content and continuation and is
 shown as unavailable, never as an empty queue or successful operation. Refresh
-is deliberate. No background request retries or write recovery occur. The
+is deliberate. A confirmed write receipt remains distinct from a failed refresh;
+minimal inert replay references do not restore protected task content. The
 profile read may perform Workstream's existing actor
-admission, last-seen and audit effects; “read-only” does not mean no server-side
-effects. This local interface does not establish a deployed service, payment,
-or future write/management workflows. See the [terminal presentation contract](../docs/design/cli-terminal-experience.md).
+admission, last-seen and audit effects. This local interface does not establish
+a deployed service, payment or complete submission/management workflows. See
+the [terminal presentation contract](../docs/design/cli-terminal-experience.md).
 
 ## Explicit commands
 
@@ -375,8 +395,9 @@ and contribution evidence with PostgreSQL, MinIO and Celery; see
 [the journey evidence](../.commitrail/initiatives/WS-CLI-001/WS-CLI-001-43.md).
 That proof is not hosted deployment verification or proof of human review.
 The [capability ledger](../docs/roadmap_status.md) identifies those remaining
-product boundaries. The interactive workspace above is still read-only;
-explicit commands remain the interface for supported mutations.
+product boundaries. The interactive workspace exposes claim/start alongside
+contributor reads and guide downloads; explicit commands remain the interface
+for preparation, submission and the other supported mutations.
 
 ### Download and install