docs: prepare agent challenge reruns

docs/add/2026-06-workflow-platform-presentation.md

@@ -171,8 +171,8 @@ uv run wf --config wf.python.config.json cap call local.ops.echo --input '{"text
 Turn it into a saved workflow:
 
 ```powershell
-uv run wf --config wf.python.config.json draft create-from-capability `
+uv run wf --config wf.python.config.json draft create `
-  python_echo_ws local.ops.echo --name python_echo
+  python_echo_ws --capability local.ops.echo --name python_echo
 
 uv run wf --config wf.python.config.json draft save python_echo_ws `
   --artifact python_echo `

docs/add/architecture.md

@@ -240,8 +240,8 @@ uv run wf --config wf.python.config.json cap call local.ops.echo --input '{"text
 Turn it into a saved workflow:
 
 ```powershell
-uv run wf --config wf.python.config.json draft create-from-capability `
+uv run wf --config wf.python.config.json draft create `
-  python_echo_ws local.ops.echo --name python_echo
+  python_echo_ws --capability local.ops.echo --name python_echo
 
 uv run wf --config wf.python.config.json draft save python_echo_ws `
   --artifact python_echo `

docs/add/system-design-implementation.md

@@ -1018,7 +1018,7 @@ The draft path demonstrates agent-oriented authoring. A draft workspace can be
 seeded from one capability's input and output schemas:
 
 ```powershell
-wf draft create-from-capability report_ws local.report.extract_report
+wf draft create report_ws --capability local.report.extract_report
 ```
 
 That command is intentionally a best-effort bootstrap, not a complete workflow
@@ -1550,13 +1550,13 @@ cap call local.report.extract_report `
 
 ## Draft Bootstrap And Focused Edits
 
-`create-from-capability` is a best-effort bootstrap. It creates a one-step
+`wf draft create --capability` is a best-effort bootstrap. It creates a one-step
 draft from the selected capability's wrapper hints. Focused commands then cover
 common edits without requiring the agent to write RFC 6902 patches by hand.
 
 ```powershell
 uv run wf --config examples/report_workflow/wf.config.json `
-draft create-from-capability report_ws local.report.extract_report `
+draft create report_ws --capability local.report.extract_report `
  --name report_case_study --title "Report Case Study"
 
 uv run wf --config examples/report_workflow/wf.config.json `

docs/add/thesis-outline.md

@@ -411,7 +411,7 @@ wf config validate
   -> wf source list
   -> wf source resources / prompts
   -> wf cap list / inspect / call
-  -> wf draft create-from-capability
+  -> wf draft create --capability
   -> wf draft save
   -> wf deploy save / validate
   -> wf run start

docs/current_roadmap.md

@@ -64,9 +64,9 @@ clear operator feedback before adding more architecture.
   compact JSON outlines for agent discovery, and emits valid self-contained
   JSON Schema with `--verbose`.
 - Completed: `wf draft bind --from ... --to ...` composes input/state/output
-  schema projection with step binding merge, replacing the narrower
+  schema projection with step binding merge, replacing the prior narrower
-  `bind-output-to-state` helper and reducing manual draft patch repairs in
+  output-to-state helper and reducing manual draft patch repairs in agent
-  agent challenge runs.
+  challenge runs.
 - Completed: draft CLI vocabulary now uses `wf draft create --capability` and
   `wf draft add-step --capability`, replacing the longer
   `*-from-capability` commands that agents repeatedly guessed around.

examples/agent_challenges/AGENTS.md

@@ -0,0 +1,14 @@
+# Agent Challenge Review Notes
+
+Before reviewing, grading, or summarizing challenge runs, read:
+
+- [`docs/runbooks/agent-challenge-evaluation.md`](../../docs/runbooks/agent-challenge-evaluation.md)
+
+Before changing or running a specific challenge, also read that challenge's
+local docs:
+
+- `browser_click_challenge/README.md`
+- `report_workflow_challenge/README.md`
+
+Keep challenge-specific instructions in each challenge's `challenge-prompt.md`.
+Keep shared harness behavior in the root `examples/agent_challenges` modules.

examples/agent_challenges/summarize_trials.py

@@ -44,7 +44,15 @@ def _dict(value: object) -> dict[str, Any]:
     return value if isinstance(value, dict) else {}
 
 
-def _attempts(agent_self_report: dict[str, Any]) -> str:
+def _attempts(agent_self_report: dict[str, Any], manual_audit: dict[str, Any]) -> str:
+    # Manual audits are authoritative for aggregate tables; the raw self-report
+    # remains visible in final-report.md for traceability.
+    audit_evidence = _dict(manual_audit.get("evidence"))
+    audit_total = audit_evidence.get("attempts_total")
+    audit_failed = audit_evidence.get("attempts_failed")
+    if isinstance(audit_total, int) and isinstance(audit_failed, int):
+        return f"{audit_failed}/{audit_total}"
+
     attempts = _dict(agent_self_report.get("attempts"))
     total = attempts.get("total")
     failed = attempts.get("failed")
@@ -86,7 +94,7 @@ def load_trial_summary(path: Path) -> TrialSummary:
         validity=_string(outcome.get("evaluation_validity")),
         duration_seconds=_float(outcome.get("duration_seconds")),
         tokens_total=_int(tokens.get("total")),
-        attempts=_attempts(self_report),
+        attempts=_attempts(self_report, manual_audit),
         read_flags=_read_flags(self_report),
         notes=" ".join(notes.split()),
     )

tests/examples/test_agent_challenge_summary.py

@@ -77,6 +77,24 @@ def test_load_trial_summary_uses_short_labels_and_manual_outcome(
     assert summary.notes == "Pass | with newline and spacing."
 
 
+def test_load_trial_summary_prefers_manual_attempt_evidence(
+    tmp_path: Path,
+) -> None:
+    from examples.agent_challenges.summarize_trials import load_trial_summary
+
+    report = _write_report(tmp_path / "results" / "trial.report.json")
+    payload = json.loads(report.read_text(encoding="utf-8"))
+    payload["manual_audit"]["evidence"] = {
+        "attempts_total": 3,
+        "attempts_failed": 2,
+    }
+    report.write_text(json.dumps(payload), encoding="utf-8")
+
+    summary = load_trial_summary(report)
+
+    assert summary.attempts == "2/3"
+
+
 def test_render_markdown_escapes_table_cells(tmp_path: Path) -> None:
     from examples.agent_challenges.summarize_trials import (
         load_trial_summary,