Files
boshu2__agentops/docs/ROADMAP.md
T
Bo 852d74b3b6 docs: reconcile legacy-RPI-lane doctrine + ADR-0009 substrate marker (ag-etgr #frozen-rpi-doctrine) (#698)
## What
Two doc-reconciliation findings from the codebase-discovery (ag-etgr P2,
ag-xa7q P3):

- **ag-etgr** — the CLAUDE.md *Legacy RPI lane* note said "do not write
new tests for"
`rpi_loop_supervisor.go`/`rpi_c2_events.go`/`rpi_phased_tmux.go`/`rpi_parallel.go`,
but **all four have substantial test suites** and the
`check-test-count-regression` ratchet expects them maintained — the
doctrine and the ratchet pulled opposite ways. Reframed to "live,
tested, **no new surface area**; extend tests for caller-driven changes;
full removal needs soc-1gbpz". Also fixed the stale "superseded by the
gc bridge" line in `docs/ROADMAP.md` (gc bridge was removed).
- **ag-xa7q** — `ADR-0009` names Gas City as the reference substrate
throughout (predates the ag-124p prune). Added a dated historical note:
the **decision stands**, but the substrate is now NTM+MCP+managed-agents
(→ docs/3.0.md). Decision record preserved for provenance.

## Note
A 3rd discovery bead (ag-xoam, BC mid-numbering) was **closed
not-reproducible** — grep showed docs already use canonical BC2
Validation/BC3 Loop/BC4 Factory consistently; the archaeology agent had
confabulated the divergent labels.

Docs-only. markdownlint clean; validate-agents-split green.

Closes-scenario: ag-etgr#frozen-rpi-doctrine
Bounded-context: BC3-Loop
Evidence: CLAUDE.md
2026-06-03 12:51:49 +00:00

2.1 KiB
Raw Permalink Blame History

Roadmap (planned, not committed)

Features designed or discussed but not yet implemented. Nothing here has a committed timeline; items may change or be dropped. If a command or capability is documented elsewhere as available, that doc is the source of truth — this page is only for what is not built yet. The 3.0 north star is docs/3.0.md.

CLI — designed, not implemented

Command Intent Status
gt convoy Convoy monitoring (track a group of related runs/worktrees) designed, not implemented
bd mol Structured workflow "molecule" pour (see bd formula/bd mol design) designed, not implemented
bd cook Batch/recipe execution over beads designed, not implemented

If you hit an error invoking one of these, it does not exist yet — that is expected.

Curation pipeline — later stages

The curation pipeline is a six-stage design: CATALOG, VERIFY, INDEX, SCORE, REJECT, CONSTRAIN. Today CATALOG and VERIFY ship as CLI commands (ao curate catalog, ao curate verify, ao curate status). The later stages (INDEX, SCORE, REJECT, CONSTRAIN) are roadmap, not built. The orthogonal, currently-active prevention lane is the finding-compiler (.agents/findings/registry.jsonl → constraints + planning rules), which is separate from this pipeline. See docs/curation-pipeline.md.

Hookless default install (ADR-0002 S2–S5)

The 3.0 north star demotes hooks to optional adapters (docs/3.0.md). The remaining lift — a default install path that requires zero hooks, and an RPI lifecycle that runs discovery→validation hookless — is tracked as ADR-0002 stages S2–S5. It is a follow-on, not part of the 3.0 reconciliation close.

Legacy RPI lane removal

The legacy RPI lane (rpi_parallel, rpi_loop_supervisor, rpi_phased_tmux + the ao rpi parallel / supervised-loop surfaces) is load-bearing but live — the phased engine kept its non-gc backends after the gc-bridge removal, and live code still references the lane's helpers. Removing it requires a caller-migration refactor, tracked as soc-1gbpz.