feat(evaluation): measure agent recovery from failed MCP calls

Makefile

@@ -57,14 +57,16 @@ e2e-test: ## Run CLI to Client SDK to Server end-to-end tests.
 	@uv run python -m pytest -n $(PYTEST_WORKERS) tests/e2e
 
 .PHONY: evaluation-unit-test
-evaluation-unit-test: ## Run the evaluation project's unit tests, the work-continuity benchmark included.
+evaluation-unit-test: ## Run the evaluation unit tests, the work-continuity and agent tool recovery suites included.
 	@uv sync --project evaluation --frozen --extra capture-ownership
 	@uv run --project evaluation --extra capture-ownership pytest -c evaluation/pyproject.toml \
 		evaluation/artifact_evolution/tests \
 		evaluation/coding/swebench_pro/tests/unit \
 		evaluation/coding/work_continuity/tests/unit \
 		evaluation/memory/longmemeval_v2/tests \
 		evaluation/memory/capture_ownership/tests -m "not live" -q
+	@# Agent tool recovery drives the real Server package, so it runs in the root project environment.
+	@uv run --frozen pytest evaluation/agent_tool_recovery/tests -q
 
 .PHONY: code-seekdb-test
 code-seekdb-test: ## Exercise native code indexing against a real embedded seekdb instance.

evaluation/README.md

@@ -6,6 +6,7 @@ Python environment and build configuration.
 
 | Directory | Purpose | Suites |
 | --- | --- | --- |
+| [`agent_tool_recovery/`](agent_tool_recovery/README.md) | Agent recovery after failed PowerContext MCP calls | [Agent tool recovery](agent_tool_recovery/README.md) |
 | [`coding/`](coding/README.md) | Coding-task completion and continuation quality | [SWE-bench Pro](coding/swebench_pro/README.md), [Work continuity](coding/work_continuity/README.md) |
 | [`memory/`](memory/README.md) | Memory retrieval, answer quality, and write-gate evidence | [LoCoMo](memory/locomo/README.md), [LoCoMo-Plus](memory/locomo_plus/README.md), [LongMemEval-V2](memory/longmemeval_v2/README.md), [Memory write gate](memory/write_gate/README.md), [Capture ownership](memory/capture_ownership/README.md) |
 | [`performance/`](performance/README.md) | Load, latency, storage growth, and capacity limits | None yet |
@@ -16,6 +17,7 @@ Python environment and build configuration.
 evaluation/
 ├── pyproject.toml                 # Shared powercontext-eval environment and build
 ├── uv.lock
+├── agent_tool_recovery/            # Agent recovery after failed MCP calls
 ├── artifact_evolution/             # Artifact-evolution recording and faithful replay
 │   ├── src/powercontext_eval_artifact_evolution/
 │   ├── tests/
@@ -53,7 +55,8 @@ evaluation/
 Run these commands from the repository root. LoCoMo and LoCoMo-Plus modules use the root project's
 Python dependencies. SWE-bench Pro, Work continuity, and LongMemEval-V2 share this directory's locked environment and
 expose independent `swebench-pro`, `work-continuity`, and `longmemeval-v2` commands. Skill-up has its own validation
-dependencies and pinned CLI. ZCode guidance uses the root project's dependencies and its pinned native CLI.
+dependencies and pinned CLI. ZCode guidance uses the root project's dependencies and its pinned native CLI. Agent
+tool recovery uses the root project's dependencies and the installed Claude Code and Codex CLIs.
 
 ```bash
 # Install the shared environment, then choose a suite command.
@@ -76,6 +79,10 @@ uv run python evaluation/skills/skill-up/validate_suite.py
 # Native ZCode Skill regression: validate the pin and runner without calling a model.
 uv run --locked python -m evaluation.zcode_guidance.pin --check
 uv run --locked python -m evaluation.zcode_guidance.run --help
+
+# Agent tool recovery: run the offline tests and inspect the real-host runner without calling a model.
+uv run pytest evaluation/agent_tool_recovery/tests
+uv run python -m evaluation.agent_tool_recovery.run --help
 ```
 
 The [SWE-bench Pro console guide](coding/swebench_pro/docs/console.md) covers its Web and worker deployment.
@@ -101,7 +108,7 @@ The `powercontext-eval` distribution contains five packages: `powercontext_eval_
 `swebench-pro`, `work-continuity`, `longmemeval-v2`, and `capture-ownership` commands and their suite modules.
 Capture ownership requires the `capture-ownership` extra shown above.
 LoCoMo behavior tests remain under [`tests/`](../tests/), including [`tests/evaluation/`](../tests/evaluation/).
-Skill-up and ZCode guidance keep their own validation tests. The [Bub replay harness](../e2e/bub/README.md) and
+Skill-up, ZCode guidance, and agent tool recovery keep their own validation tests. The [Bub replay harness](../e2e/bub/README.md) and
 [product E2E tests](../tests/e2e/) cover cross-component acceptance.
 
 Keep fixed datasets, locks, configuration examples, and scoring rules with their owning suite. Store local outputs in