feat: add profiled agent challenge harness

examples/agent_challenges/profile-prompts/all.md

@@ -0,0 +1,5 @@
+## Instruction Profile: all
+
+Start with the supplied skills and public docs. If genuinely blocked, you may
+inspect broader repository docs, examples, tests, and source. Report what you
+read; observed tool calls will also be retained for audit.

examples/agent_challenges/profile-prompts/none.md

@@ -0,0 +1,7 @@
+## Instruction Profile: none
+
+Use challenge files, `wf --help`, `wf schema`, validation, inspect, and bounded
+trace commands. Do not read repository skills, docs, examples, tests, source,
+prior trials, or prior stores. If public surfaces are insufficient, report the
+exact blocker and finish the task as failed rather than reverse-engineering the
+implementation.

examples/agent_challenges/profile-prompts/skills.md

@@ -0,0 +1,6 @@
+## Instruction Profile: skills
+
+Use the supplied skills under `.agent/skills/` plus public `wf` commands. Do not
+read repository examples, tests, source, prior trials, or prior stores. If the
+skills and public surfaces are insufficient, report the exact blocker rather
+than reverse-engineering implementation code.