ur-agent 1.80.2 → 1.80.4

package/CHANGELOG.md

@@ -1,5 +1,48 @@
 # Changelog
 
+## 1.80.4
+
+- Stopped runaway WebFetch loops on permanent HTTP 4xx responses. Failures are
+  keyed by normalized URL rather than the hidden summarization prompt, so
+  changing prompts or alternating between two dead links cannot bypass the
+  circuit breaker. The first repeated URL is refused without another network
+  request; continued repetition stops the turn.
+- WebFetch now distinguishes permanent client errors from transient timeouts,
+  conflicts, early-data responses, rate limits, and server failures. Permanent
+  errors direct the agent to WebSearch, a parent/index page, or another source;
+  transient failures remain retryable. URL identities are hashed, bounded,
+  scoped to the active query, and cleared when that query completes.
+- Plan mode now explicitly creates visible implementation tasks with
+  `TaskCreate` before workspace changes. Task-gate recovery clarifies that
+  entering or updating a plan is not a substitute for the task list. Approve
+  All remains unchanged.
+- Completed shell-command display hardening in the background-task detail
+  dialog. Tabs, control bytes, invisible characters, and bidirectional controls
+  are made explicit before width truncation, matching Bash and PowerShell.
+- Unified prompt and standalone configuration assignments. `ur config set`
+  now supports accessibility, thinking, output, compaction, and Vim/editor
+  settings; `ur config get` and `ur config list` expose effective safe values
+  for automation without hand-editing JSON.
+- Added `ur a2a card --v1` to preview the existing current A2A 1.0 Agent Card,
+  plus `ur session status [session-id]` to inspect an explicit conversation or
+  the latest resumable conversation for the current project.
+
+## 1.80.3
+
+- Restored visible task planning with a strict-hybrid default. Atomic,
+  low-risk requests still execute directly; requests with multiple outcomes,
+  sequencing, plan mode, delegation, project-sized work, or release/security/
+  migration/production risk must create an actionable task before the first
+  mutation. Read-only investigation remains unrestricted.
+- The task gate now uses deterministic user-turn classification instead of
+  treating conversation length as complexity. It fails closed when task state
+  is unreadable, rechecks permission-rewritten tool input, preserves active
+  interrupted boards, and never deadlocks a custom tool profile that omits
+  `TaskCreate`.
+- Operators can keep task tracking advisory with
+  `tasks.requireBeforeChanges.enabled=false`, or set `freeReads=0` to require a
+  task before every mutation. Approve All remains unchanged.
+
 ## 1.80.2
 
 - Repaired repeated `AskUserQuestion` validation loops when a model emits one

package/README.md

@@ -81,6 +81,13 @@ useful. Each task records id, order, title, description, status, dependencies,
 assigned logical agent role, input, expected output, verification criteria, file
 targets, risk level, and whether approval is required.
 
+Inside the interactive agent, task tracking uses a strict-hybrid default:
+atomic low-risk work stays direct, while multi-outcome, sequenced, planned,
+delegated, project-sized, and high-risk lifecycle work must create a visible
+task before the first mutation. Reads stay unrestricted. Configure the gate
+with `tasks.requireBeforeChanges`; `enabled=false` makes it advisory and
+`freeReads=0` requires a task before every mutation.
+
 During real `ur exec` runs the task board streams when a task status changes,
 then appears again in the final report. Quiet/non-interactive runs can suppress
 streaming while preserving the final board. Public status labels are
@@ -437,7 +444,7 @@ as first-class subcommands in the shipped CLI.
 | `ur auth claude` | Launch the official Claude Code login flow. |
 | `ur auth gemini` | Use the official Gemini CLI login flow where supported. |
 | `ur auth antigravity` | Use the official Antigravity CLI login flow where supported. |
-| `ur config set` | Persist safe non-secret provider settings such as provider, model, base URL, command path, and fallback. |
+| `ur config set|get|list` | Set or inspect validated non-secret agent, accessibility, editor, and provider settings. |
 | `ur mcp` | Configure Model Context Protocol servers, expose fail-closed built-in tools over standard input/output, or run the secure stateless web server with Tasks and Apps. |
 | `ur skill` | Initialize, run, strictly verify, Ed25519-sign, and create trusted signing keys for portable skills. |
 | `ur ag-ui serve` | Start the secure AG-UI HTTP/SSE adapter for user-facing applications with truthful capability discovery. |
@@ -446,7 +453,7 @@ as first-class subcommands in the shipped CLI.
 | `ur acp` | Connect editors through the Agent Client Protocol, or manage the separate UR HTTP JSON-RPC server. |
 | `ur exec` | Run one or more prompts in non-interactive mode with optional concurrency. |
 | `ur ide diff` | Capture editor-readable inline diff bundles. |
-| `ur a2a card` | Print Agent-to-Agent discovery metadata. |
+| `ur a2a card` | Print legacy Agent-to-Agent discovery metadata; add `--v1` to preview the current A2A 1.0 card. |
 | `ur a2a serve` | Start the negotiated Agent-to-Agent server with automatic client compatibility. |
 | `ur sdk` | Show programmatic headless usage and scaffold SDK examples. |
 | `ur trigger` | Parse a GitHub/Slack webhook payload and optionally launch a headless UR run. |