| name | tooling-scout-autonomy |
| description | Build leg of the scout family - what novel software to BUILD on your existing toolset, ranked for a long autonomous run, biased to refactors and wide backfills. Triggers - autonomy scout, scout autonomy, what should I build. |
scout-autonomy
The build leg of the scout family. Its two siblings move tooling in and out:
tooling-scout-capability finds what to add
(external catalog in), tooling-scout-displacement
finds what to shed (your custom code out). This leg answers the third question:
given the toolset you already have, what novel software should you build on top of it?
Add / shed / build. The output is a ranked, buildable backlog plus a ready-to-run long-run
contract, sized to hand straight to an overnight autonomous session.
Triggers
autonomy scout, scout autonomy, what should I build, build leg, plan a long run, overnight build plan.
The bias that defines this leg
Novel does not mean net-new. The ranking is deliberately weighted so that:
- Refactors and progressive improvements of existing subsystems outrank greenfield.
- Wide backfills - author or extend one thing, roll it across N repos (a new pre-commit validator, a config migration, a doc-trifecta sweep) - rank highest, because reach is the dominant term.
- Net-new must clear a higher bar: it only ranks if it composes at least one tool you already maintain. Standalone greenfield is demoted by default.
This is leverage scouting, not idea generation. Detail in references/scoring-and-config.md.
Phases
Five phases, each checkpointing to a single-day notes/scratch file so a run resumes
without re-running phase 1. Inbox path: <notes-dir>/YYYY-MM-DD-scout-autonomy-{phase}.md.
Outputs go to notes/scratch, not the personal-OS repo - speculative projection
surfaces private intent, and only an actual build lands commits.
- Phase 1 - Grounded sweep - walk the working surface, ingest sibling-scout and backlog outputs if present, list buildable candidates tagged refactor / backfill / net-new.
- Phase 2 - Projection - project software that does not exist yet but composes the existing toolset. Composition is required, not optional.
- Phase 3 - Leverage scoring - score value x reach / cost with the refactor + backfill multipliers, apply the reserved-surface fence, rank.
- Phase 4 - Contract synthesis - for the top picks, write a long-run contract (goal / done-condition / non-goals) ready for an overnight run.
- Phase 5 - Present and hand off - flatten the top 3 to chat, explicit-deny approval, the chosen contract seeds a long-run session.
Run config
Supplied at invoke, not hardcoded (this is a public-safe method): target repo list,
notes/scratch path, and a reserved-surface fence - subsystems this run must not touch
or propose work on, the operator's way to keep a scout off reserved or boundary code.
Schema in references/scoring-and-config.md.
Paired with
The scout family is one portfolio activity in three directions:
- scout-capability - catalog in, acquire what you lack.
- scout-displacement - inventory in, retire what OSS already does.
- scout-autonomy (this skill) - toolset in, build the high-leverage thing on top.
See also
