loopflow 0.12.28

Run steps and flows with coding agents
Documentation
---
requires: requested scope and available evidence
produces: a sourced judgment, useful authorized action or explicit no-action result
action_style: procedural
---
Advance useful delivery while preserving each unit’s autonomy.

Use the requested scope: a Wave and its Tasks, or the repository and its Waves.
When `LOOPFLOW_FLOW_NAME` is `vsm-operate`, default to the invocation repository
even with an ambient Task/Wave; only an explicit request narrows it. Read earlier
pass findings. For repository scope, use `lf wave list --current --json`, then
`lf wave status <wave> --json` and `lf roadmap --wave <wave> --json` for the selected
Waves. Report missing scope; never substitute another repository or Wave.

Read the intended outcomes and actual delivery evidence of the Tasks or Waves
in scope. Follow the work to its beneficiary: agent activity, a green PR or a
closed Task alone does not prove usefulness.

Find the next useful move the responsible unit can carry itself. Inspect
`lf task status <issue>` before starting selected, authorized work with
`lf task run <issue>`; leave active work and waiting reviews intact. Prefer
removing a concrete impediment over taking the unit's work into a central queue.
Unclear coordination goes forward to S2; capacity tensions to S3. Neither is a
prerequisite for otherwise independent progress.

Keep dated sources and distinguish observations from hypotheses. Stale or failed
reads leave unknowns; they cannot authorize closure or a competing launch. Act
within authority, preserving independent Task progress and review gates. A no-action
result is valid; do not manufacture work or reports.

Reply with findings, observed action results and open questions. In a Flow, reuse
one `scratch/` note keyed by `LF_FLOW_STEP`'s `invocation` across S1–S5 for scope,
evidence and disagreements. Durable changes belong in the existing plan, Task or
memory. Confirm authorized writes; drafts and failed deliveries remain pending.