CHANGELOG.md

@@ -86,6 +86,21 @@ The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
   This eliminates redundant Alembic migrations across all test suites and reduces
   total test suite wall-clock time. (#2334)
 
+- **CI — Nox output captured as CI artifacts**: All 8 nox-running CI jobs
+  (`lint`, `typecheck`, `security`, `quality`, `unit_tests`, `integration_tests`,
+  `e2e_tests`, `coverage`) now pipe their output through `tee` to named log files
+  and upload them as Forgejo artifacts (`ci-logs-<job>`) with `if: always()` so
+  logs are available even on failure. Multi-session jobs append to a single combined
+  log. Artifacts are retained for 30 days. Seven agent definition files updated to
+  reference the artifact download API for CI diagnostics. (#2782)
+
+- **Benchmarks — ASV performance benchmark suite for providers module**: Added 5
+  new ASV benchmark files under `benchmarks/` covering all performance-sensitive
+  paths in the `providers` module: `ProviderCostTable` construction and iteration,
+  `CostTracker` accumulation, `FallbackSelector` selection, `ProviderRegistry`
+  enumeration and lookup, and LLM adapter instantiation for all 5 built-in
+  providers. 68 benchmark methods total. (#3022)
+
 ### Fixed
 
 - Fixed session leak in all `AutomationProfileRepository` public methods
@@ -160,6 +175,24 @@ The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 - `ThoughtBlockWidget` background corrected from `$primary 20%` to `$primary-muted 20%`,
   making thought blocks visually lighter and more subtle per spec §29811. (#1448)
 
+- **CLI — `agents plan list` gains `--namespace`/`-n` filter**: The `plan list`
+  command now accepts `--namespace`/`-n` to filter plans by namespace, bringing it
+  into alignment with the specification and achieving parity with `agents action list`.
+  The service layer already supported namespace filtering; only the CLI layer was
+  missing the option. The Filters panel in rich output displays `Namespace: <value>`
+  when the flag is active. (#2616)
+
+- **MCP — Error message extraction corrected for MCP 1.4.0**: `MCPToolAdapter.invoke()`
+  now reads error messages from `content[0].text` per the MCP 1.4.0 protocol
+  specification, replacing the previous non-standard `error` key lookup that silently
+  returned `"unknown error"` for all real MCP 1.4.0 servers. Safe guards prevent
+  `IndexError` on absent or malformed `content` arrays. (#2600)
+
+- **CI — All quality gates restored to passing on master**: Resolved format
+  violations, incomplete A2A JSON-RPC 2.0 migration in test step files, stale
+  session CLI output format assertions, plan list output format regressions, and
+  stale plan apply test data. No quality gates were suppressed or bypassed. (#2629)
+
 ## [3.7.0] — 2026-04-02
 
 ### Added

docs/api/mcp.md

@@ -68,6 +68,27 @@ await adapter.disconnect()
 | `invoke(name, arguments)` | Call a tool with input validation |
 | `register_tools(registry)` | Bulk-register discovered tools |
 
+### Error Handling in `invoke()`
+
+`MCPToolAdapter.invoke()` complies with the **MCP 1.4.0** protocol for error
+responses.  When a tool call returns `isError: true`, the error message is
+extracted from `content[0].text`:
+
+```python
+# MCP 1.4.0 error response shape
+{
+    "isError": True,
+    "content": [{"type": "text", "text": "Tool execution failed: ..."}]
+}
+```
+
+Safe guards prevent `IndexError` when `content` is absent or empty — the
+adapter falls back to `"unknown error"` in those cases.
+
+> **Note (v3.7.0+):** Prior to this fix, `invoke()` read from a non-standard
+> `error` key that no real MCP 1.4.0 server populates, causing all error
+> messages to silently appear as `"unknown error"`. (#2600)
+
 ---
 
 ## `MCPToolDescriptor`

docs/reference/plan_cli.md

@@ -109,6 +109,7 @@ agents plan list [REGEX] [OPTIONS]
 
 | Flag                    | Description                                      |
 |-------------------------|--------------------------------------------------|
+| `--namespace`, `-n`    | Filter by namespace                               |
 | `--phase`              | Filter by phase (strategize, execute, apply)      |
 | `--state`              | Filter by processing state                        |
 | `--processing-state`   | Alias for `--state`                               |
@@ -116,6 +117,22 @@ agents plan list [REGEX] [OPTIONS]
 | `--action`             | Filter by action name                             |
 | `--format`, `-f`      | Output format                                     |
 
+### Examples
+
+```bash
+# Filter by namespace
+agents plan list --namespace myteam
+
+# Short form
+agents plan list -n myteam
+
+# Combined namespace + state filter
+agents plan list --namespace myteam --state processing
+
+# Namespace filter with no results
+agents plan list --namespace nonexistent
+```
+
 ### Rich Output
 
 The `rich` format renders a **Plans** table with columns:
@@ -131,8 +148,9 @@ The `rich` format renders a **Plans** table with columns:
 
 After the table:
 
-- **Filters** panel — shown only when at least one filter (`--phase`, `--state`,
-  `--project`, `--action`) is active. Lists the active filter values.
+- **Filters** panel — shown only when at least one filter (`--namespace`, `--phase`,
+  `--state`, `--project`, `--action`) is active. Lists the active filter values,
+  with `Namespace: <value>` displayed first when `--namespace` is provided.
 - **Summary** panel — total plans, processing count, completed count, and errored count.
 
 Followed by a `✓ OK N plan(s) listed` success message.