rnd-calibratelisted

# R&D Framework: Calibrate Invoke skill: `rnd-framework:rnd-calibration` ## Step 1: Parse Arguments Parse `$ARGUMENTS` by splitting on whitespace: - **`session-id`** — first token (e.g. `20260316-154145-1227`) - **`task-id`** — second token (e.g. `T3`) - **`true-verdict`** — third token: `PASS` or `FAIL` - **`reason`** — remainder of the string (optional free-text) If any of `session-id`, `task-id`, or `true-verdict` are missing, use `AskUserQuestion`: > "Please provide: session-id, task-id, and true-verdict (PASS or FAIL)." ## Step 2: Read the Original Verdict Get the project base directory and locate the verification report: ```bash BASE_DIR=$("${CLAUDE_PLUGIN_ROOT}/lib/rnd-dir.sh" --base) VERDICT_FILE="${BASE_DIR}/sessions/<session-id>/verifications/<task-id>-verification.md" ``` Note: `--base` returns the branch-scoped base dir. Sessions are always located inside the branch-scoped dir. Read `$VERDICT_FILE`. If it does not exist, display an error: > "No verification report found at `$VERDICT_FILE`. Check the session-id and task-id." Extract the verdict line (e.g. `## Verdict: PASS`) from the report to show the original verdict alongside the correction. ## Step 3: Append Correction Record Determine the calibration file path. Use `CLAUDE_PLUGIN_DATA` if set; fall back to the slug-root path via `--calibration` otherwise: ```bash CALIB_FILE="${CLAUDE_PLUGIN_DATA:-$("${CLAUDE_PLUGIN_ROOT}/lib/rnd-dir.sh" --calibration)}" ``` `--calibration` returns the un-parti