← ClaudeAtlas

rnd-verifylisted

Run criticality-conditional verification on a built task. Verifies against pre-registered criteria with information barrier.
oleksify/rnd-framework · ★ 0 · AI & Automation · score 75
Install: claude install-skill oleksify/rnd-framework
# R&D Framework: Verify Determine the RND artifacts directory: ```bash RND_DIR=$("${CLAUDE_PLUGIN_ROOT}/lib/rnd-dir.sh") ``` Read the plan from `$RND_DIR/protocol.md` (and the assertions from `$RND_DIR/validation-contract.md`). Check `TaskList` to confirm which tasks are built and ready for verification. ## CRITICAL: Information Barrier Enforcement You MUST NOT read `$RND_DIR/builds/T<id>-self-assessment.md` files during verification. The `read-gate.sh` hook blocks these reads mechanically. You wrote the self-assessment during the build phase, but during verification you must assess work purely against the pre-registered spec. ### Pre-Flight Check Run this sanity check **once** before verification: ```bash SA_FILES=$(ls "$RND_DIR/builds/"*self-assessment* 2>/dev/null || true) if [ -n "$SA_FILES" ]; then echo "INFO-BARRIER: The following files must NOT be read during verification:" echo "$SA_FILES" fi ``` ## Execution Use `TaskUpdate` to mark target tasks as `in_progress` before verifying. If $ARGUMENTS is a task ID: verify that one task. If $ARGUMENTS is a wave: verify all tasks in the wave. If $ARGUMENTS is "all": find all built but unverified tasks. **Wave-batched spawn:** Spawn one Verifier agent per wave (not per task). The Verifier receives all task pre-registrations for the wave and returns a `wave-N-verdict-map.json` plus per-task artifacts. Criticality drives the per-agent model boost (see `rnd-framework:rnd-orchestration`); there is no parallel-judge