feat: unify Entity skill control plane#1
Draft
StaticObserver wants to merge 13 commits into
Draft
Conversation
…ol plane Remove the v3 Case state machine, v4 deterministic flow facade, and all compatibility commands (flow, writer, project, inspect, run-status, migrate --from-v3) now that the live controller has been imported into the v5 SQLite store. - add entityctl site add/list so v5 can register Sites without the deleted file-based profile tooling; migrate --export becomes export - trim entity_router_common to the v5/shared primitive set - pgen preflight reads the v5 store; managed writes fail closed until a v5 pgen Goal exists - drop v3/v4 templates, playbooks, and the v4-era evals/router-flow scaffolding; rewrite router references and contract tests - sync README and skill docs; record the retirement in the v5 implementation design doc
…evidence Checkpoint of the pre_gold formal evaluation scaffold (submission/physics oracles, schemas, fake Slurm, variants), the router-operation evidence validator, and CI coverage before trimming to the lightweight A/B setup.
Keep only task.md, physics-spec.json, and RUNBOOK.md. The pre_gold formal scaffold (oracles, schemas, fake Slurm, variants) stays recoverable in git history. Drop the scaffold-specific test and CI steps, and document per-round directory conventions in the runbook.
- 0.1.0: R1-R4 review fixes; executor sbatch uses -input; apply failure finishes anomaly + operation cancel; status --live degrades to UNKNOWN; env-build requires OpenMPI >= 5.0.0 (hard gate); VERSION file + store migrate shell - 0.2.0: entityctl site discover + site-profile schema; submission create/verify contract with body fingerprint; structured scheduler remediation hints - 0.3.0: parameter confirmation gates across the chain - env-build checkpoint confirm, pgen preflight card/confirm, router planner simulation-confirmation hard gate (digest + signer, hard fail) - 0.4.0: goal_kind on plans with per-kind validation; build Goal (checkpoint compatibility=pass + parameter digest, double-checked at plan and apply); data Goal with executor data.inventory.v1 receipts; readiness/current generalized over asset dimensions - evals: e2e-neutral-streaming review findings, harness review, oracle tests, observability phase tracing - design: router development plan 2026-07-22 (0.x versioning, 1.0.0 candidate criteria) Tests: 121 passed, 3 subtests passed
- status --live now classifies out-of-band divergence: job_gone (with sacct-confirmed flag), state_mismatch (recorded submitted vs observed terminal state), untracked_job (foreign job in the Case run root); bounded to three scheduler queries, tolerant of missing sacct - entityctl apply --refresh re-executes a completed data Goal plan so a stale data-inventory.json is rebuilt after run artifacts change; data.inventory.v1 always re-walks the run root instead of trusting a previous verified receipt - doctor: hard failures (exit 2) on client bundle drift and invalid stored Site profiles; warnings for leaked active Operations with the exact operation cancel remedy - skill adoption metrics in the phases report: skill vs raw invocation counts, sqlite control-plane surgery counter, skill_call_share — the clean-comparison metric required by the 1.0.0 candidate criteria - contract test updated: live status budget is now three bounded scheduler queries Tests: 132 passed, 3 subtests passed
…lity, pgen router: stable plan_hash across replans, timezone-safe launch recovery with sacct fallback, purge directory probe fix, sqlite error envelope, ssh_alias whitelist, content-addressed build identity, single-transaction cancel, claim-token write checks, goal.schema for run/build/data env-build: whitelist versions/arch/extra_env keys (command injection and path traversal), fix ENTITY_DEPS_ROOT derivation, cmake_cxx_flags double quoting, precise override issue removal, structured snapshot failures, run_id validation, reject unknown Kokkos arch observability: protected-root check for phases --output, case-insensitive path compare, real kimi isError flags, robust evidence validation, id-matched acceptance evidence, span_id reuse rejection, fcntl fallback, TOCTOU-safe finish/artifact, redaction fixes pgen/nt2py: preflight always emits JSON contract, store_present field, fsync decisions, new nt2py unit tests docs: sync README/CONTRIBUTING/design contract with implementation Adds 61 regression tests (193 total, all passing), each reproduced the reported failure before the fix.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
entityctlfacade, shared project-to-Case bindings, and content-addressed multi-client bundle publicationWhy
Codex, Claude Code, and Kimi Code could each maintain overlapping skill and project-state copies, while ordinary Router work required long model-driven primitive loops. The new control plane gives all local providers one controller-authoritative state location that remains readable when execution sites are offline, and moves routine validation and orchestration into bounded deterministic programs.
User and developer impact
Validation
python3 -m unittest discover -s tests -v— 84 passedpython3 -m unittest discover -s skills/entity-pgen/tests -v— 3 passedpython3 -m unittest discover -s skills/entity-env-build/tests -v— 18 passedquick_validate.pygit diff --cached --checkpassedScope boundary
Real SSH build/run/nt2 runner staging and a matched-scenario cross-provider token-reduction canary remain follow-up work. The diagnostic Claude/Kimi sessions in this PR are explicitly not treated as matched comparisons.