phuc-nt Strip Identity
Rewrites every upstream identity marker in kit content to the my-agent-kit equivalents, then verifies none remain. Run this after every upstream sync, before committing or publishing.
When to use
- After bulk-copying upstream content into
claude/(upstream sync workflow) - Before
git commit/npm publishfollowing any upstream merge - As a CI gate —
verify.shexits non-zero if any marker leaks
Why this skill exists
The v2.0.0 release was stripped by hand with sed + \b word boundaries.
BSD/macOS sed does not honor \b consistently, so the ck:<skill> slash
command pattern slipped through — 246 files shipped with /ck:cook instead
of /mk:cook (fixed in v2.0.1). These scripts encode the verified
boundary-safe patterns so that bug class cannot recur.
The two scripts
| Script | Purpose | Exit |
|---|---|---|
scripts/strip.sh [dir] |
Apply all substitutions in place | 0 = applied |
scripts/verify.sh [dir] |
Fail if any marker remains | 0 = clean, 1 = dirty |
Default target is claude/. Always run verify.sh after strip.sh.
bash claude/skills/strip-identity/scripts/strip.sh
bash claude/skills/strip-identity/scripts/verify.sh # must exit 0
What gets rewritten
| Marker | Becomes |
|---|---|
claudekit-engineer |
my-agent-kit |
claudekit-marketing |
my-marketing-kit |
claudekit-cli / claudekit-docs |
my-kit-cli / my-kit-docs |
ClaudeKit / CLAUDEKIT / claudekit |
MyKit / MYKIT / my-kit |
docs.claudekit.cc / claudekit.cc |
docs.my-kit.local / my-kit.local |
github.com/claudekit/... |
(removed) |
@goonnguyen / mrgoonie / goonie |
personal |
Udit Goenka |
original author |
.ck.json / .ckignore |
.mk.json / .mkignore |
ck-config / ck-paths |
mk-config / mk-paths |
ck:<name> (slash command) |
mk:<name> |
ckm:<name> (marketing command) |
mk:<name> |
`ck` / spawn('ck') / ck <subcommand> (bare CLI) |
mk equivalents |
Critical implementation notes
The ck: rule. A naive s/ck:/mk:/g corrupts check:, block:,
track:, feedback:, stack:, lock:. The safe pattern is a negative
lookbehind — replace ck: only when not preceded by an ASCII letter:
(?<![a-zA-Z])ck:
This matches /ck:fix, ck:cook, "ck:test" but never the inside of a
longer word.
The ckm: and bare-ck rules (added after the 2026-08 A/B test). Two
marker classes escaped every pattern above and shipped for months:
ckm:design (the string ckm: does not contain ck:), and bare CLI
mentions like requires `ck` CLI or spawn('ck') (no colon, no dash —
nothing for the prefix rules to anchor on). The bare form is the worst kind
of leak: agents read it and try to run a binary that does not exist. strip.sh
now rewrites the unambiguous forms (backticked, quoted-as-command, and
ck <known-subcommand>); verify.sh scans \bck\b and flags any
surviving standalone ck for manual review rather than guessing.
Perl, not grep -P or sed. macOS/BSD grep has no -P (PCRE); BSD sed
has no lookbehind. Both scripts use perl, whose regex engine is identical
across platforms. perl is a hard dependency — the scripts abort if missing.
Substitution order matters. Longest project names (claudekit-engineer)
are rewritten before the bare claudekit, otherwise claudekit consumes the
-engineer/-cli suffix first and leaves my-kit-engineer.
metadata.json is excluded from the ck-paths / .ck checks. Its
deletions array intentionally retains legacy paths (e.g.
hooks/lib/ck-paths.cjs) so mk-init.sh --upgrade knows to delete them from
old installs. Do not "fix" those entries.
Adding a new marker
Add the substitution to both scripts:
strip.sh— a news/.../.../gline (mind the ordering rule above)verify.sh— extend the relevantscan '...'pattern
Then re-run the fixture test (see git history of this skill for the test
fixture used to validate the ck: lookbehind against check:/block:).