feat(ase-code-analyze): add -Q one-shot analyze → resolve → implement → verify

[Brusdeylins]

Summary

Adds a --quick/-Q flag to ase-code-analyze that turns an analysis
into a fully autonomous one-shot analyze → resolve → implement → verify
pipeline. After reporting and persisting the surviving problems, -Q:

• groups them into per-file clusters (same-file problems are never
  resolved in separate parallel worktrees, which would guarantee
  conflicts);
• dispatches one worktree-isolated ase-code-resolve -Q sub-agent
  per cluster in the background, emitting a progress line per
  returning cluster; each worktree is first seeded with the live
  tree's uncommitted state so its returned diff applies cleanly;
• reconciles the diffs into the working tree (plain git apply,
  --3way fallback, and a conflicting cluster is re-resolved
  sequentially so that all problems merge), then removes the temporary
  worktrees again;
• runs the project's formatter, build, and test suite centrally on
  the merged result, with up to two autonomous repair rounds.

ACCEPTED findings (documented, reviewed trade-offs) are excluded
from auto-resolution. Nothing is staged or committed — staging stays
with the user. -Q composes with --severity (only surviving findings
are resolved) and requires a Git repository (otherwise clusters are
resolved sequentially, without worktree isolation).

Scope

Additive and fully guarded: the whole flow lives inside a single
<if <getopt-option-quick/> …> block, so the default (non--Q)
behavior is unchanged apart from the argument-hint, allowed-tools
(Bash), and getopt-spec additions. Only ase-code-analyze
(SKILL.md + help.md) is touched (+189/−2), rebased cleanly on top of
the current severity-descending ordering.

Verification

Field-tested end-to-end on a real, multi-finding Java/Gradle codebase.
The current design — per-file clustering, worktree seeding, --3way
reconcile, background fan-out, central verify-and-repair, and ACCEPTED
exclusion — is the direct result of hardening that first run surfaced
(unverified per-worktree fixes, dirty-tree patch failures, same-file
conflicts, and silent long fan-out).

🤖 Generated with Claude Code

[Brusdeylins]

Closing this without merge: the underlying need is solved better by orchestration than by a flag.

The -Q one-shot pipeline (analyze → resolve → implement → verify, with per-file clusters, worktree isolation, reconciliation and a central verification pass) encoded a fixed multi-step workflow inside a single skill and its option parser. With the ase-meta-workflow skill now shipping upstream, that same pipeline is expressible as an explicit, inspectable orchestration of the existing skills - without widening ase-code-analyze's surface, without duplicating the resolve/implement logic, and without a hard-wired fan-out policy.

So the branch is dropped rather than rebased further. Nothing here is needed by the other open PRs: ase-code-resolve --quick|-Q exists upstream independently, and the --mode all forwarding in #65 stands on its own.

🤖 Generated with Claude Code