0.60.0

Added

• aep plan artifact new review-result … --findings <file|-> takes the findings as a JSON array, validated against the entry schema and written into the body as a JSON findings block. A body that also opens a block is refused as ambiguous. (#38)

Fixed

• Planning writer fence files no longer land in the working tree. Inside a Git repository they live under <git-common-dir>/aep/, shared by every linked worktree; outside one they stay beside the store. While an older aep may still hold the in-tree names, this build also locks them when they exist, without creating them. (#37)
• A malformed findings block is refused at one body line, with that line quoted and a hint naming JSON as the machine-written form. Whatever --findings accepts reads back unchanged, and a findings fence quoted inside a longer fence is prose rather than the review's findings. (#38)
• A render, store rebuild, write or validate refused after staging the projection no longer leaves its .engineering/planning.aep-stage-* directory beside the projection. 0.59.2 removed the stage on one refusal only; every refusal between staging and publishing now removes it. Recovery from a killed process's committed watermark still publishes.
• A planning.aep-stage-* directory left by a process killed during publication is removed by the next publication. A stage is removed only when its lock is free and its pid runs nothing in this pid namespace; a stage a live process holds, in this pid namespace or another, a name that is not exactly a stage's, and a stage whose owner cannot be established are left alone. Staging no longer clears a directory already at its own name, which with two pid namespaces sharing one checkout could be another live process's stage.