Inspect Yarli state, durably enqueue new tranches, and choose `yarli run continue` versus `yarli run --fresh-from-tranches` for long feedback loops.
Use this skill when the user wants an agent to monitor Yarli, queue newly discovered work durably, and keep execution moving without losing state.
.yarli/tranches.toml.yarli run, yarli run continue, or yarli run --fresh-from-tranches.key, summary, allowed_paths, verify, done_when, and optional max_tokens.yarli run continue only for work already owned by the continuation snapshot and only when no drift is present.yarli run --fresh-from-tranches after new tranche enqueue or other live-plan changes.Run the helper script:
bash "$SKILL_DIR/scripts/yarli-loop-inspect.sh" [project-root]
Review:
.yarli/continuation.json.yarli/tranches.tomlyarli run status and yarli run explain-exit output when a run ID is availableUse the enqueue wrapper instead of ad hoc prose:
bash "$SKILL_DIR/scripts/yarli-enqueue-tranche.sh" \
--project-root <project-root> \
--key TP-05 \
--summary "Implement config loader hardening" \
--allowed-paths crates/yarli-cli/src \
--verify "cargo test -p yarli --bin yarli cli_parses_" \
--done-when "fresh-versus-continue workflow is documented and tested"
This wrapper always uses yarli plan tranche add --idempotent.
If the tranche key already exists with different effective fields, stop and report the mismatch instead of silently mutating scope.
yarli run.yarli run continue.yarli run --fresh-from-tranches.--prompt-file <path> to yarli run or yarli run --fresh-from-tranches.Do not end a cycle with an unstructured refusal and no next step.
references/codex.md.references/claude.md.End every cycle with this plain-text block:
YARLI_DECISION_V1