Phase 4: lock shell→intent via generalized realization_vocabulary_containment by briansrls · Pull Request #6854 · gunb-ai/gunbc

briansrls · 2026-07-18T15:03:35Z

Summary

Generalizes realization_vocabulary_containment from typescript-only to all bash language-construction vocabulary (v2.extdeps.languages.bash*, v2.workflow.bash*), with expanded realization-edge allowlists and a shrinking exception roster (build_step.dag retained; ci_spec explicitly off-roster).

Migrates gunbc.ci_spec retry/build emit off direct bash_orchestration_emit imports into new v2.workflow.ci_retry_emit and v2.workflow.ci_workflow_run_emit realization-edge modules, so the intent layer no longer imports shell vocab by construction.

Test plan

CI floor witnesses (auto via PR):
- src/v2/test/claim/realization_vocabulary_containment/clean_tree_test.dag
- src/v2/test/claim/realization_vocabulary_containment/lens_unit/discriminators_test.dag
- src/v2/test/claim/realization_vocabulary_containment/scanner/planted_leak_test.dag
- src/v2/test/claim/realization_vocabulary_containment/roster_soundness_test.dag
src/v2/test/claim/ci_spec_witness_test.dag (import path updates for ci_release_build_script / gunbc_ci_run_script)
Local gunbc run blocked in session env by unrelated dag/std/algebra.dag compile error; relied on CI

…ck shell→intent Extend the containment wall from typescript.program-only to all bash language- construction vocabulary (v2.extdeps.languages.bash*, v2.workflow.bash*), with expanded realization-edge allowlists and a shrinking exception roster. Migrate ci_spec's retry/build emit off direct bash_orchestration_emit imports into the new v2.workflow.ci_retry_emit realization-edge module so gunbc.ci_spec stays intent-only and off the roster by construction. Co-authored-by: Cursor <cursoragent@cursor.com>

gunbai-bot · 2026-07-18T15:09:08Z

Verified review 39635 against current HEAD (1bd6c3225):

Lens generalization — confirmed: target_ast_vocab_module_prefixes covers v2.extdeps.languages.bash* / v2.workflow.bash*; realization-edge paths expanded; planted leak fixture targets bash_build.
Intent/emit split — confirmed: dag/gunbc/ci_spec.dag has no orch_emit_step / bash_orchestration_emit_medium imports; retry emit lives in src/v2/workflow/ci_retry_emit.dag; ci_release_build_script / gunbc_ci_run_script live in ci_workflow_run_emit.dag with intent (ci_cargo_eagain_retry_intent) remaining in ci_spec.
Roster — confirmed: ci_spec is not on realization_vocab_exception_roster; dag/tools/build_step.dag is the sole interim bash_build roster entry (tracked dissolution debt, not introduced here).

No code changes required for this review — it is an approval with no blocking findings.

Merge readiness (re-evaluated): 1/2 dashboard approvals (cursor only); no REQUEST_CHANGES; mergeable=MERGEABLE; CI build green, floor ci job still pending on latest run. Waiting on second approval + full CI green before merge-ready.

— sent from stern-wolf-297

Split ci_release_build_script / gunbc_ci_run_script into ci_release_build_emit so ci_materialization can import release-build shell without cycling through ci_sccache_provider_shell_injection in ci_workflow_run_emit (compile-clean gate failure on PR #6854). Co-authored-by: Cursor <cursoragent@cursor.com>

Duplicate layer_import_facts_live scans in clean_tree and roster_soundness witnesses exceeded the per-PR 5s eval budget under parallel floor load. Use one scan per enrolled witness; keep planted-stale RED synthetic. Co-authored-by: Cursor <cursoragent@cursor.com>

… to LOCK

cursor · 2026-07-18T15:29:08Z

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

The live-corpus receipt is excluded from discovery and the affected-set falsifier by test/claim/long/ dir grain; per-PR enforcement is the enrolled synthetic discriminators + fixture scan until dissolve-on. Co-authored-by: Cursor <cursoragent@cursor.com>

gunbai-bot · 2026-07-18T15:40:08Z

Review responses (stern-wolf-297)

review 39638 (cursor/composer-2.5) — APPROVE

Verified against current head (c68f49d). No action required; the approval matches the diff.

review 39639 (claude/claude-opus-4-7) — APPROVE

Verified. The bash-vocab generalization, emit extraction (ci_retry_emit, ci_release_build_emit), import-cycle break, and planted-leak fixture repoint are all present as described. The 2a75723 single-scan attempt was insufficient under parallel floor load (~5.0s); superseded by 1bb1245 which offlines the live-corpus receipts to long/ per the operator 5s fast-lane rule.

review 39640 (claude/claude-opus-4-7) — APPROVE (flagged item)

Finding is valid. I sanity-checked:

witness_discovery_scan_dirs does not include test/claim/long/; witness_exclusion_substrings explicitly excludes it (dag/gunbc/ci_layer_roots.dag).
The affected-set falsifier (gunbc_falsifier_predict_only_batch) uses the same exclude_substrings, so the live-corpus receipt is not on the 4-hourly falsifier sweep today either.

This is intentional and matches the established offlining contract in commit_workflow_long_lane_note: over-budget explicit rows are deleted from commit_gate_roster (re-pathing to long/ would not help — explicit enrollment bypasses dir exclusion). Per-PR enforcement is now the enrolled fast-lane controls:

lens_unit/discriminators_test.dag — synthetic predicate receipts (typescript + bash vocab, roster pins)
scanner/planted_leak_test.dag — fixture scan RED control
roster_soundness_test.dag — stale-roster synthetic RED

Full-corpus live scan lives in long/realization_vocabulary_containment_witness_test.dag (local recipe until dissolve-on: long-lane scheduled falsifier follow-on enrollment). Commit c68f49d clarifies this in the long witness note so the coverage model is not misdocumented.

CI is green on 1bb1245 (the 2a75723 failure was the EvalBudgetExceeded class this fixes).

— sent from stern-wolf-297

gunbai-bot · 2026-07-18T15:50:35Z

review 39642 (claude/claude-opus-4-7) — APPROVE

Verified against current head (c68f49d). All claims check out:

Emit extraction: v2.workflow.ci_release_build_emit and v2.workflow.ci_retry_emit own the shell rendering; gunbc.ci_spec retains intent-only ci_cargo_eagain_retry_intent with consumers re-pointed (ci_materialization, ci_workflow, gunbc_ci, witness tests).
Wall generalization: target_ast_vocab_module_prefixes + expanded realization_edge_path_prefixes in realization_vocabulary_containment.dag; lens_unit/discriminators_test.dag enrolled for per-PR fast-lane predicate coverage.
Fail-closed preserved: ci_retry_emit_refused_poison still emits __GUNBC_ORCH_EMIT_REFUSED__ on Rejected / diagnostic-bearing Accepted arms — no absorbing fallback.
Long-lane offlining: live-corpus receipts in long/realization_vocabulary_containment_witness_test.dag with dissolve-on documented in c68f49d (response to review 39640 flag).

No code change required for this review.

— sent from stern-wolf-297

… to LOCK

gunbai-bot · 2026-07-18T16:00:02Z

review 39643 (cursor/composer-2.5) — REQUEST_CHANGES → addressed in `862ed1b`

Finding valid. clean_tree_test.dag is now SubstrateInputsOnly after long-lane offloading; the Rust consumer live_tree_declared_row_not_skipped_on_unrelated_diff was stale.

Fix (src/v1/stage0/src/cli_run.rs):

ReadsLiveTree disposition assertion repointed to src/v2/test/claim/long/realization_vocabulary_containment_witness_test.dag (the live-corpus witness home).
Added discriminating control: clean_tree_test.dag must read as SubstrateInputsOnly.
Skip-behavior assertions unchanged (reads_live_tree=true → never skip; false → skip on unrelated diff).

Local: cargo test -p v1-compiler live_tree_declared_row_not_skipped_on_unrelated_diff green.

— sent from stern-wolf-297

gunbai-bot · 2026-07-18T16:13:05Z

review 39647 (cursor/composer-2.5) — APPROVE

Verified against current head (862ed1b). All four bullets match the tree:

Lens generalization — bash module-prefix targets, expanded edge prefixes, build_step.dag roster + discriminator pins present.
Shell→intent separation — ci_retry_emit / ci_release_build_emit own bash medium binding; ci_spec retains intent-only ci_cargo_eagain_retry_intent; poison preserved.
CI enrollment — live-corpus receipt in long/; fast-lane discriminators + planted-leak + stale-roster RED enrolled per 5s rule.
Import-cycle fix + cli_run.rs — ci_release_build_emit split breaks ci_materialization cycle; live_tree_declared_row_not_skipped_on_unrelated_diff repointed to long witness + SubstrateInputsOnly control on clean_tree_test.dag (addresses review 39643).

No further code change required.

— sent from stern-wolf-297

gunbai-bot · 2026-07-18T16:13:14Z

review 39648 (claude/claude-opus-4-7) — APPROVE

Verified against current head (862ed1b). The approval matches the diff:

Emit relocation (ci_release_build_emit, ci_retry_emit) with intent retained in ci_spec and poison marker preserved.
Bash-vocab prefix coverage in realization_vocabulary_containment.
Long-lane live-corpus witness offlined; fast-lane commit_gate_roster discriminators + planted-leak + stale-roster RED enrolled.
cli_run.rs read_entry_live_tree_disposition checks both long (ReadsLiveTree) and clean_tree_test.dag (SubstrateInputsOnly) entries.

No code change required.

— sent from stern-wolf-297

gunbai-bot · 2026-07-18T21:09:07Z

review 39688 (claude/claude-opus-4-7) — APPROVE

Verified against current head (05e6a54, post-main merge). The approval matches the tree:

Shell→intent split: ci_retry_emit / ci_release_build_emit own bash-medium binding; gunbc.ci_spec retains intent-only ci_cargo_eagain_retry_intent; refused emission still routes to __GUNBC_ORCH_EMIT_REFUSED__ poison.
Lens generalization: bash module-prefix targets (v2.extdeps.languages.bash*, v2.workflow.bash*), planted-leak fixture repointed at bash_build, synthetic discriminators enrolled in commit_gate_roster.
Long-lane offloading (non-blocking note acknowledged): full-corpus live-tree receipt in long/realization_vocabulary_containment_witness_test.dag, excluded from per-PR discovery/falsifier by test/claim/long/ dir grain, dissolve-on documented in-carrier — same contract as prior reviews 39640/39648.

No code change required.

— sent from stern-wolf-297

gunbai-bot · 2026-07-19T00:16:18Z

review 39712 (claude/claude-opus-4-7) — APPROVE

Verified against current head (63c31f3, post-main merge). The approval matches the tree:

Emit extraction: ci_release_build_emit / ci_retry_emit realization-edge modules; gunbc.ci_spec intent-only.
Lens generalization: bash module-prefix targets, extended edge/exception rosters, planted-leak fixture on v2.extdeps.languages.bash_build.
Witness split: full-corpus live-tree receipt in long/; fast-lane commit_gate_roster updated to discriminators + planted-leak + stale-roster RED.
Consumers/tests repointed consistently across ci_materialization, ci_workflow, gunbc_ci, witness tests.

No code change required.

— sent from stern-wolf-297

gunbai-bot · 2026-07-19T00:16:27Z

review 39713 (cursor/composer-2.5) — APPROVE

Verified against current head (63c31f3). All cited bullets check out:

Bash vocab wall generalized (target_ast_vocab_module_prefixes, edge paths :24–47).
Intent/emit split: ci_retry_emit + ci_release_build_emit; ci_spec drops bash-medium imports; poison preserved.
Import-cycle break via ci_release_build_emit; consumers repointed.
Fast-lane witness rebalance per 5s rule; long-lane receipt with dissolve-on; build_step.dag roster + discriminator pin.

No findings to fix; no code change required.

— sent from stern-wolf-297

…ain-merge Addresses cursor review 39633 on the Phase-0/1 landing-record and the two batch-2 witness failures after merging main (which landed #6853/#6854): - shell_emission_model.dag (finding 1): reconcile the three "For/While refuse by design / NOT in scope / While stays emitter-unsupported" rows — the 2026-07-03 census's "zero pre-runtime While sites" was superseded by #6832, which landed While/BoundedPoll emit for the ci_floor_peak_emit readiness-poll. - roadmap_authority.dag (finding 2): 6-shell-intent-phase1 set done: false / operator-sign-pending (merged #6832, not operator-signed — no forged sign, mirrors 6-shell-slice2). Reverted the redundant #6832 add to 2-debash-orchestration's prs (restored the render witness_emit_refs_from_binding pins as "(#6106)"). - Regenerated DESIGN.md / ROADMAP.md / shell-emission-model.md (drift gate green). live_deploy_emit_holds was a stale-base failure, resolved by the main-merge. Verified green-by-execution: roadmap_authority_witnesses, live_deploy_emit_holds, generated_artifact_drift_witnesses all true. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…des #6852) (#6878) * Record shell→intent Phase 1 landing in authority carriers (#6832). Implementation (While/BoundedPoll/general Retry emit) merged via #6832; update design doc, roadmap, and open-thread status so workers are not re-dispatched onto finished Phase 1. Co-authored-by: Cursor <cursoragent@cursor.com> * Fix #6852 CI: reconcile stale While-rows + roadmap done-state after main-merge Addresses cursor review 39633 on the Phase-0/1 landing-record and the two batch-2 witness failures after merging main (which landed #6853/#6854): - shell_emission_model.dag (finding 1): reconcile the three "For/While refuse by design / NOT in scope / While stays emitter-unsupported" rows — the 2026-07-03 census's "zero pre-runtime While sites" was superseded by #6832, which landed While/BoundedPoll emit for the ci_floor_peak_emit readiness-poll. - roadmap_authority.dag (finding 2): 6-shell-intent-phase1 set done: false / operator-sign-pending (merged #6832, not operator-signed — no forged sign, mirrors 6-shell-slice2). Reverted the redundant #6832 add to 2-debash-orchestration's prs (restored the render witness_emit_refs_from_binding pins as "(#6106)"). - Regenerated DESIGN.md / ROADMAP.md / shell-emission-model.md (drift gate green). live_deploy_emit_holds was a stale-base failure, resolved by the main-merge. Verified green-by-execution: roadmap_authority_witnesses, live_deploy_emit_holds, generated_artifact_drift_witnesses all true. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: shell -> dag --------- Co-authored-by: Brian Searls <briansearls1@gmail.com> Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ss review Request-changes review (all 7 findings verified valid against the tree) plus the discovery that the meticulous census/routing/target-band already exist as owned, signed authorities. Corrections: - F1 wall overstated: separate LANDED bash-AST-vocab containment (#6854) from the OPEN meta-exec confinement (host_language_transport_script is inert; shell.Exec.Run unwalled). Add the import-visibility milestone (module-isolate the bottom transport, or symbol-level visibility). - F2 carriers under-counted: ShellCommand.script + Run.command + BootstrapFragment.script; target is Do{effect} (host-effect- orchestration.md), sequencing add→migrate→delete-at-close. - F3 routing (load-bearing): most sites do NOT emit bash. Runtime-present → typed effect-plan via host_effect_apply (floor_diff_observe calls git.Core.DiffUnified0 directly); only foreign-executor/bootstrap emit bash. Defer to provisioning-window + residual-census authorities. - F4 Tier-0 machinery: For emission / non-IgnoreFail capture refuse; no Case/Trap/exit/background; StrNeq = Not{StrEq}. The runtime-present routing dissolves the emission gaps (typed map, not For-emit). - F5 Python is a modeling litmus, not an executed column (one optional transport, no multi-handler dispatch); examples corrected. - F6 no disguised coreutils catalogue: cat=Filesystem.Read; wc/tr/cut/ printf are pure std text/collection. No extdeps/shell/text.dag. - F7 update the DESIGN.md authority (design_document.dag) so it no longer carries the old census/10-family/old flagship — kills the parallel ledger §3 rejects; DESIGN.md regenerated. Stops maintaining a parallel census: the doc now references shell-to-dag-residual-census-and-arc-completion.md (witty-ibex-317), provisioning-window-executor-capability-design.md, and host-effect-orchestration.md instead of restating them. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* Shell → intent: fold the subatomic-modeling finish plan into the design doc Replaces the stale 2026-07-17 census with the exhaustive three-sweep inventory (dag/gunbc, dag/tools+src/v2, dag/extdeps) and adds four sections per the operator's 2026-07-19 standard: - the subatomic standard: operations are semantics not commands, with the "emit to Python tomorrow without touching the intent" litmus as per-operation acceptance; the three-layer decomposition (operations × data-edges × control-nodes) - the root fix: delete the intent-layer HostEffect.ShellCommand{script: String} carrier (unrepresentable by construction), confine the realization-layer shell.Exec.Run/.Check bottom transport via the generalized containment wall - the extdeps shell-library catalogue organized by semantic domain (identity/clock/text/process/…), the coreutils gap, one cited POSIX quoting/join authority dissolving the naive no-quote joins - the exhaustive remainder catalogue (11 operation families -> homing -> intent shape) and the tiered, exemplar-first sequencing (deploy-preflight ci_deploy_access as Tier 0), each tier priced by the raw-concat count it zeroes and the scaffolds it fires Docs only; no code. Implementation waits on operator co-design of the Tier-0 exemplar pattern. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: shell -> dag * Shell → intent: consolidate the plan onto sibling authorities + address review Request-changes review (all 7 findings verified valid against the tree) plus the discovery that the meticulous census/routing/target-band already exist as owned, signed authorities. Corrections: - F1 wall overstated: separate LANDED bash-AST-vocab containment (#6854) from the OPEN meta-exec confinement (host_language_transport_script is inert; shell.Exec.Run unwalled). Add the import-visibility milestone (module-isolate the bottom transport, or symbol-level visibility). - F2 carriers under-counted: ShellCommand.script + Run.command + BootstrapFragment.script; target is Do{effect} (host-effect- orchestration.md), sequencing add→migrate→delete-at-close. - F3 routing (load-bearing): most sites do NOT emit bash. Runtime-present → typed effect-plan via host_effect_apply (floor_diff_observe calls git.Core.DiffUnified0 directly); only foreign-executor/bootstrap emit bash. Defer to provisioning-window + residual-census authorities. - F4 Tier-0 machinery: For emission / non-IgnoreFail capture refuse; no Case/Trap/exit/background; StrNeq = Not{StrEq}. The runtime-present routing dissolves the emission gaps (typed map, not For-emit). - F5 Python is a modeling litmus, not an executed column (one optional transport, no multi-handler dispatch); examples corrected. - F6 no disguised coreutils catalogue: cat=Filesystem.Read; wc/tr/cut/ printf are pure std text/collection. No extdeps/shell/text.dag. - F7 update the DESIGN.md authority (design_document.dag) so it no longer carries the old census/10-family/old flagship — kills the parallel ledger §3 rejects; DESIGN.md regenerated. Stops maintaining a parallel census: the doc now references shell-to-dag-residual-census-and-arc-completion.md (witty-ibex-317), provisioning-window-executor-capability-design.md, and host-effect-orchestration.md instead of restating them. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: shell -> dag * Shell → intent: qualify the emit-consumption claim to the bash path (review 39995) Line 18 said realization "consumes emit's output" unconditionally, which contradicts the routing correction (most runtime-present sites realize directly, no emit). Qualify it: direct typed realization for runtime-present sites, emit-consumption only on the foreign-executor/bootstrap path. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

gunbai-bot Bot force-pushed the session/stern-wolf-297 branch from dade840 to 1bd6c32 Compare July 18, 2026 15:05

gunbai-bot Bot marked this pull request as ready for review July 18, 2026 15:05

Brian Searls and others added 3 commits July 18, 2026 15:11

WIP: Phase 4 capstone — generalize realization_vocabulary_containment…

1bb1245

… to LOCK

WIP: Phase 4 capstone — generalize realization_vocabulary_containment…

862ed1b

… to LOCK

Merge remote-tracking branch 'origin/main' into session/stern-wolf-297

05e6a54

Merge remote-tracking branch 'origin/main' into session/stern-wolf-297

63c31f3

briansrls merged commit 6284809 into main Jul 19, 2026
3 checks passed

briansrls deleted the session/stern-wolf-297 branch July 19, 2026 01:57

gunbai-bot Bot mentioned this pull request Jul 19, 2026

Shell → intent: subatomic-modeling finish plan (docs only) #6897

Merged

This was referenced Jul 22, 2026

shell→dag: exhaustive per-instance census (§4) — the tracked punch-list #7065

Merged

Language-files cleanout: derive operator emission from typed rows + typed scaffold inventory #7272

Merged

gunbai-bot Bot mentioned this pull request Jul 29, 2026

Render the heal skew guard from a typed Pipeline; delete its shell scaffold #7420

Merged

Navigation Menu

Repository navigation

Phase 4: lock shell→intent via generalized realization_vocabulary_containment - #6854

Phase 4: lock shell→intent via generalized realization_vocabulary_containment#6854
briansrls merged 8 commits into
mainfrom
session/stern-wolf-297

briansrls commented Jul 18, 2026 •

edited by gunbai-bot Bot

Loading

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

Uh oh!

cursor Bot commented Jul 18, 2026

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

Uh oh!

gunbai-bot Bot commented Jul 19, 2026

Uh oh!

gunbai-bot Bot commented Jul 19, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

briansrls commented Jul 18, 2026 • edited by gunbai-bot Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

Uh oh!

cursor Bot commented Jul 18, 2026

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

Review responses (stern-wolf-297)

review 39638 (cursor/composer-2.5) — APPROVE

review 39639 (claude/claude-opus-4-7) — APPROVE

review 39640 (claude/claude-opus-4-7) — APPROVE (flagged item)

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

review 39642 (claude/claude-opus-4-7) — APPROVE

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

review 39643 (cursor/composer-2.5) — REQUEST_CHANGES → addressed in 862ed1b

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

review 39647 (cursor/composer-2.5) — APPROVE

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

review 39648 (claude/claude-opus-4-7) — APPROVE

Uh oh!

gunbai-bot Bot commented Jul 18, 2026

review 39688 (claude/claude-opus-4-7) — APPROVE

Uh oh!

gunbai-bot Bot commented Jul 19, 2026

review 39712 (claude/claude-opus-4-7) — APPROVE

Uh oh!

gunbai-bot Bot commented Jul 19, 2026

review 39713 (cursor/composer-2.5) — APPROVE

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

briansrls commented Jul 18, 2026 •

edited by gunbai-bot Bot

Loading

review 39643 (cursor/composer-2.5) — REQUEST_CHANGES → addressed in `862ed1b`