Rename the harden skill to improve

[laird]

Renames the platform-neutral validation & refinement skill harden → improve. It was never security hardening — it's a run → grade → fix → log loop for stabilizing a platform, so improve reads truer.

Changes

• skills/harden/ → skills/improve/ (git mv)
• SKILL.md: name, title, description, and example verbs updated to improve; dropped the now-redundant "(not security hardening)" disclaimer
• README.md descriptive mention and skills/autocoder/references/model-config.md → improve / skills/improve/
• Left the dated 3.32.0 changelog entry as-is (it accurately records the skill shipped as harden); a rename note belongs with the release that carries this
• No mirrors: the skill lived only in top-level skills/ (not codex/factory)

Drift checker (check-optional-skills-drift.sh) green.

⚠️ Two things to flag

1. Pre-existing CI failure, not from this change. tests/test_worker_launch_lib.sh fails on origin/master too (a Gemini worker-command expectation: expected '/fix-loop', got 'bash …/gemini-fix-loop.sh'). This PR touches only skills/ + docs — no scripts. If the shell check is red here, that's the base-branch failure, not this diff.
2. Targets master on purpose. The skill only exists on master, and integration is currently 38 commits behind master (other work merged straight to master), so it can't host this rename cleanly. See the note below.

Branching note (needs a decision)

The feature → integration → master discipline set up earlier has drifted: master advanced to 3.33.0/autocoder 4.11.0 via direct merges while integration sits at 3.29.0, so the earlier release PR #91 is now stale. Happy to reconcile — merge master into integration (or reset integration to master) and re-open the release flow — just say which.

🤖 Generated with Claude Code

---

Generated by Claude Code

[laird]

CI blocker is pre-existing, not from this PR. The shell check is red on exactly one test:

Failing shell tests (1):
  tests/test_worker_launch_lib.sh (exit 1)
FAIL: Gemini worker command: expected '/fix-loop', got 'bash '.../scripts/gemini-fix-loop.sh''

This reproduces on a clean origin/master checkout — it predates this change, which is a docs‑only skill rename (skills/ + README + model-config.md, no scripts). pytest is green and every other shell test passes. So this red X is the base‑branch failure, not a regression in the rename.

I'm not fixing it here (unrelated scope, and it's ambiguous whether the test expectation or worker-launch-lib.sh is the stale side). Happy to fix it as its own PR if wanted.

---

Generated by Claude Code

[laird]

Superseded by #112 which was merged moments ago — same harden→improve rename with the same model-tier and description updates.