This release adds 3 notable features for engineering teams evaluating rollout.
Published 20d
Developer Productivity
✓ No known CVEs patched
✓ No known CVEs patched in this version
Topics
ai-coding-assistant
archetype
ast-analysis
claude-code
claude-plugin
code-conventions
+12 more
code-review
code-style
developer-tools
javascript
linter
mcp
mcp-server
prompt-engineering
python
ruby
rails
typescript
Summary
AI summaryDeep‑work now hires experts for parallel subagents, adds verification steps, and improves worktree handling.
Full changelog
Added
- Deep-work hires experts and manages its own worktree defensively. The
skill's dig is now staffed, not solo: a "Hire experts (dispatch discipline)"
section makes parallel subagents the default posture whenever two or more
independent unknowns exist - read-only code scouts, version-pinned web
researchers, and a fresh-context reviewer at verification - each dispatched
with one owned question, the context it cannot discover alone, and a
required answer shape. An expert's answer is input, not truth: every claim
a decision rests on is verified first-hand before it enters the brief (new
brief checkbox), and hired agents return evidence only - the brief decides.
Step 3's external research gains a depth ladder (pinned-version docs, then
the changelog across the exact version window, then the installed package's
own source; a blog post is a lead, never a citation). Step 5's worktree
setup was rewritten against empirically tested git behavior (both
corrections refuted the guidance it was adapted from, live on git 2.50.1):
detect an existing linked worktree by comparing
git rev-parse --path-format=absolute --git-dirwith the common dir (the
raw-output comparison false-positives from a subdirectory, and a plain
submodule does NOT produce the mismatch), use an existing worktree only
when it is dedicated to the task, honor a user-declared placement above any
tool default, never edit the user's.gitignoreto make a placement work,
and run a dependency-install + gate baseline before the first edit so
Step 6 attributes new failures (inherited ones are reported, not fixed).
Worktree infeasibility is triaged as a rule-2c hard dependency at Step 1,
and the failure report covers the no-worktree-created case.
(skills/chameleon-deep-work/SKILL.md)
Note: the /chameleon-deep-work skill itself and the docs overhaul shipped in
v2.55.0; this release refines that skill.
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About Chameleon
All releases →Beta — feedback welcome: [email protected]