Skip to content

feat(graph): add full_verify preset — one-command build->spec->test->review - #130

Merged
alpibrupa merged 1 commit into
mainfrom
full-verify-preset
Sep 5, 2026
Merged

feat(graph): add full_verify preset — one-command build->spec->test->review#130
alpibrupa merged 1 commit into
mainfrom
full-verify-preset

Conversation

@alpibrupa

Copy link
Copy Markdown
Contributor

Summary

  • The chaining mechanism already existed generically via `from_spec` (`--pipeline=build,spec,test,review` already worked) — what was missing was a dedicated, memorable preset for exactly this sequence, matching Promote the build->spec->test->review graph.lex pipeline to one-command 'full verify' #120's own title.
  • The closest existing preset, `impl_then_spec_then_test`, runs test and review in parallel after spec, so review sees the same untested output test is concurrently checking rather than reviewing what test verified.
  • Adds `full_verify()` = build → spec → test → review as a strict sequence, wired into `preset()`/`preset_names()`/`preset_shape()`/`is_preset()`. Extended `spec_matches_preset()` to assert `build,spec,test,review` and `full_verify` produce identical shapes, matching the existing discipline for the other three presets.

Fixes #120

Test plan

  • `lex check`/`lex fmt --check` — ok
  • Full repo sweep — all pass
  • `lex test` — 4/4 passed
  • `lex doc-sync --check` — current
  • Live end-to-end verification with `run_graph_persistent` against a real task ("implement fn double"): spec genuinely located and read the file build wrote (via glob/read tool calls), wrote and proved 4 real spec properties via `lex_spec_check` (self-correcting two real syntax mistakes along the way), test recognized the existing `examples{}` coverage instead of duplicating it, and review used `effects_of`/`attestation_query`/`lex_check` for a real, evidence-based verdict — confirmed this is genuine chaining, not four independent runs sharing a task description.
  • Persistent trail inspection: all four `.lex/sessions/{impl,spec,test,review}.db` trails exist with distinct, mode-appropriate event counts — spec's alone has 4 real `verified.spec_check` attestations (matching the 4 proved properties) and 2 `cap.failed` events (matching the two self-corrected mistakes); test and review each have their own `verified.type_check`.

🤖 Generated with Claude Code

…iew (#120)

The chaining mechanism already existed generically via from_spec (any
comma-separated agent list, e.g. --pipeline=build,spec,test,review
already worked) — what was missing was a dedicated, memorable preset
for exactly this sequence, matching #120's own title. The closest
existing preset, impl_then_spec_then_test, runs test and review in
PARALLEL after spec, so review sees the same untested output test is
concurrently checking rather than reviewing what test verified.

Adds full_verify() = build → spec → test → review as a strict
sequence, wired into preset()/preset_names()/preset_shape()/is_preset().
Extended spec_matches_preset() to assert build,spec,test,review and
full_verify produce identical shapes, matching the existing discipline
for the other three presets.

Live-verified end to end (not just type-checked) with
run_graph_persistent against a real task: spec genuinely located and
read the file build wrote (via glob/read tool calls), wrote and proved
4 real spec properties via lex_spec_check (self-correcting two real
syntax mistakes along the way), test recognized the existing examples{}
coverage instead of duplicating it, and review used effects_of/
attestation_query/lex_check for a real, evidence-based verdict — not
four independent runs that happened to share a task description.

Inspected all four persistent trails (.lex/sessions/{impl,spec,test,
review}.db) afterward: each has distinct, mode-appropriate event
counts and kinds — spec's trail alone has 4 real verified.spec_check
attestations (matching the 4 proved properties) and 2 cap.failed
events (matching the two self-corrected mistakes); test and review
each have their own verified.type_check. Every stage's activity is
attributable to that stage specifically.

Fixes #120

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@alpibrupa
alpibrupa merged commit 6168cb6 into main Sep 5, 2026
1 check passed
@alpibrupa
alpibrupa deleted the full-verify-preset branch September 5, 2026 09:18
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Promote the build->spec->test->review graph.lex pipeline to one-command 'full verify'

1 participant