Aee Gate Skill

AEE Gate

Apply the AEE exit-code contract to an existing rich assessment.

User input

$ARGUMENTS

Resolve assessment=<path> inside .specify/extensions/aee/assessments/ unless an explicit in-project path is supplied. Refuse symlinks and run:

python .specify/extensions/aee/scripts/python/run_aee.py gate --input "<assessment>"

To gate the most recent assessment for a lifecycle phase instead of naming a file, use the pointer that assess maintains:

python .specify/extensions/aee/scripts/python/run_aee.py gate --latest "<phase>"

--latest refuses (exit 2) when no assessment has been recorded for the phase, and prints the resolved assessment's age to stderr — check it before trusting the verdict, because it gates whatever the pointer names, however old. --input and --latest are mutually exclusive.

Exit semantics are: 0 for pass or warn, 1 for iterate, clarify, or gather_evidence, and 2 for block or malformed input.

To require lane 1 (machine verification) to pass on its own facts — regardless of where lane 2 (human review and acceptance) stands — add --require-lane1-pass:

python .specify/extensions/aee/scripts/python/run_aee.py gate --input "<assessment>" --require-lane1-pass

The runner forwards the option to the engine's gate unchanged and preserves its exit codes exactly: 0 when lane 1 passes, 1 when lane 1 is in the iterate family (recovery needed), and 2 when the engine blocks or refuses on integrity grounds (for example, a lane-1 outcome recorded in the file that the recorded facts do not support). Use it when a step must know the machine checks passed even though people are still reviewing; it says nothing about lane 2, and it never replaces the headline gate for a step that needs both lanes done. For multi-evaluator release policy, use /speckit-evaluator-report format=gate on a composed result.

A gate verdict is a lane-1 (machine verification) decision: it says the recorded evidence passed the recorded checks — in the normal path, acceptance is exactly this, the frozen requirement tests passing and recorded as a test verdict, not a human act. It does not say any human has explained, accepted, or digested the claims — read the assessment's lane-2 state (lanes.lane_2) and its not_understood entries alongside the gate result before treating gated work as settled. A gated assessment that lacks a not_understood statement is itself incomplete and the gate refuses it (exit 2).