[antoinetoussaint-byte]

Why

#1207 merged at ad957aee with no review. The independent review at that commit (/tmp/codex-cli1207-review.md) reproduced two material defects:

1. Module filter runs before full-graph validation. codefly list runnables --module other --env staging reports valid DOCUMENTS__… runnable-scope selections as unanswered_keys, while the unfiltered run of the same environment reports none. A selection the environment answers correctly is called unanswered because the check was scoped to the filtered module.
2. --tools drops a runnable whose Policy.Tool is nil, although the selected core's ToolFromPrepared already synthesises the default (ApplyText / EFFECT_MUTATION). Core's contract is that every runnable is a tool; the tests added in feat(list): runnables carry the tool they are offered as, and each one's grant #1207 encode the old absence instead.

The fix shape is known — reuse ToolFromPrepared and validate the full graph before filtering, not second defaults — but that is new code with its own review, and it is not on blue's path. v0.1.220 therefore ships without #1207; it re-lands fixed in v0.1.221 under its owner.

What this is

A plain git revert -m 1 ad957aee of the merge commit: cmd/list/runnables.go, cmd/generate/runnable_bindings.go, pkg/runnables/{derived,grant,projection}.go, their tests and docs/{cli-reference,commands}.md return to their state before #1207. Nothing else moved. Not merged until the release coordinator says so.

Evidence

go build ./...; go test ./cmd/list/ ./cmd/generate/ ./pkg/runnables/ ./cmd/ green (the CLI reference gate included, since docs/cli-reference.md is reverted with the help text); golangci-lint run clean on the reverted packages.

🤖 Generated with Claude Code