This repository was archived by the owner on Jul 18, 2026. It is now read-only.
Repository navigation
Releases
Releases · legout/pi-execflow
v1.7.12
- Fixed TDD ship/work chains so
validation-fixloops inside the chain
(--loop 5) before review/finalization, and tightened/validation-fixso
a fix-applying pass emitsGate: PASSwhen the repaired state validates.
This prevents a completed repair from leaving onlyGate: REVISEevidence
and immediately falling through to a non-closing finalizer. - Moved persisted validation gates to
.pi/execflow-validation-gate.json, with
read fallback for legacy.execflow/validation-gate.json, so repos that track
.execflow/plansdo not see the gate as a delivery artifact. - Taught review/finalize guidance to treat execflow runtime state files,
including validation gates, as unrelated workflow state rather than ambiguous
work-item changes. - Added package validation to lock the TDD chain declarations to
validation-fix --loop 5.
v1.7.7
Changes
- Fixed
/ef-autoship-tddready-queue ordering for review follow-ups by requiring created follow-up work items to depend on the reviewed original item. - Prevents autoship from implementing review follow-ups inside a dirty tree before the parent work item has been finalized and committed.
- Added package validation for the review follow-up dependency guidance and output status.
Validation
npm run validate-packagegit diff --check
v1.7.5
Fixes
- Fixed model selection so each ship-chain step switches to its configured primary model instead of sticking on the fast selector model. The fast selector models (
gpt-5.4-mini,kimi-for-coding) are now exclusive to the fast role, and/ef-updatemigratesplan/implementationdefaults that previously shared them.
v1.7.4
Fixes
- Fixed fast work-item selector prompts to prefer Codex mini with Kimi fallback, and taught
/ef-updateto migrate projects that still have the old defaultfastmodel. This prevents/ef-ship*and/ef-autoship*chains from stalling when the Zai fast endpoint is unavailable.
v1.7.3
v1.7.2
v1.7.1
Fixed
- Fixed
/ef-ship*and/ef-autoship*ready-work draining so productive next-ready dispatch iterations write a local convergence marker, while no-ready stop iterations remain no-op and converge cleanly. - Documented and validated the autoship convergence marker used to bridge prompt-template write/edit-based loop convergence with delegated subagent execution.
v1.7.0
Added
- Added the
ef-finalizersubagent for fresh-context evidence checks, dirty-tree classification, final notes, commits, and safe tracker closure. - Added a machine-readable review
# Finalization Handoffblock so downstream finalization can distinguish clean closure, delegated follow-ups, and unresolved review findings.
Changed
- Wired
/finalizeto theef-finalizersubagent so mixed delegated/direct ship chains end on finalization output instead of replaying stale review output after closure. - Hardened finalization policy around explicit evidence, review handoff parsing, dirty-tree classification, and related-file-only commits before tracker closure.
- Added package validation for the
ef-finalizeragent template.
v1.4.0
- Added deterministic tracker-tool preflight to /init-execflow so br and tk modes fail before scaffolding when their required CLI is missing.
- Added br/bv agent-instruction guidance so native tool blocks are managed in the project-root AGENTS.md instead of being duplicated in .execflow/AGENTS.md.
- Documented tk help as the local task-management entrypoint for generated .execflow/AGENTS.md files.
v1.3.0
1.3.0 - 2026-06-14
- Changed
/ef-autoshipand/ef-autoship-tddfrom nestedrun-promptorchestration prompts into chain-loop wrappers that no longer require/prompt-tool on. - Added
ship-resolveandship-tdd-resolveselector prompts so/ef-ship,/ef-ship-tdd, and autoship wrappers can select readybrissues ortktickets directly. - Added no-target and
--nextready-work draining to/ef-shipand/ef-ship-tdd, with retry tracking through.execflow/autoship-progress.json. - Added
tk readysupport to the autoship state helper while preserving configured tracker preference and ready order. - Tightened review/finalize guidance so work-item review findings can be captured as follow-up work before conservative finalization closes the original item.