[aitrawog]

Retirement evidence — executed, not asserted

The references this PR removes belong to repositories that are provably retired at source. Both facts are read live:

$ gh api -H 'Accept: application/vnd.github.raw' repos/opencharly/layer-charly-openclaw/contents/README.md?ref=main | head -1
# layer-charly-openclaw — RETIRED
$ gh api -H 'Accept: application/vnd.github.raw' repos/opencharly/layer-charly-openclaw/contents/charly.yml?ref=main | grep -cE '^\s+[a-z0-9-]+:(\s*$|\s)'
0
$ gh api -H 'Accept: application/vnd.github.raw' repos/opencharly/layer-openclaw-full/contents/README.md?ref=main | head -1
# layer-openclaw-full — RETIRED
$ gh api -H 'Accept: application/vnd.github.raw' repos/opencharly/layer-openclaw-full/contents/charly.yml?ref=main | grep -cE '^\s+[a-z0-9-]+:(\s*$|\s)'
0
$ gh api -H 'Accept: application/vnd.github.raw' repos/opencharly/layer-openclaw-full-ml/contents/README.md?ref=main | head -1
# layer-openclaw-full-ml — RETIRED
$ gh api -H 'Accept: application/vnd.github.raw' repos/opencharly/layer-openclaw-full-ml/contents/charly.yml?ref=main | grep -cE '^\s+[a-z0-9-]+:(\s*$|\s)'
0

Each of those repositories carries a RETIRED tombstone README and zero entity declarations in its charly.yml — the entities are gone from the live graph, so a reference to one is genuinely stale.

One retirement is NOT yet complete, and this PR does not claim otherwise: opencharly/pod-openclaw reads # pod-openclaw with 42 entity declarations and no tombstone. Its retirement is in flight in another session's open PRs (pod-openclaw#8, #9 — branch feat/retire-family*), which this session does not touch. This diff contains no pod-openclaw reference.

How tested

The R5 claim-keyed sweep, executed on this branch and pasted with its output:

$ grep -cE 'openclaw-(desktop|full|full-layer|full-ml-layer)' charly.yml
0

$ git show origin/main:charly.yml | grep -cE 'openclaw-(desktop|full|full-layer|full-ml-layer)'
9

So the changed file goes from 9 matching lines on main to 0 on this branch. The org-required validate / validate check is the merge gate.

Scope note — what this PR does not claim

This PR changes only the lines this diff changes. It does not claim the refs-list cleanup is complete: main's refs list still names the retired repos, and the open opencharly/marketplace#434 is the PR that removes them. The retirement of the repositories themselves is already fact, as proved above.

Rulebook compliance

• R0 - skills first: /charly-internals:git-workflow before the branch and PR actions.
• R1 - an earlier revision asserted liveness without executed evidence and left the skill documenting a state it could not prove; the evidence above is executed and the one incomplete retirement is named rather than glossed.
• R2 - not a split. The remaining surface is carried by the named open marketplace#434, tracked by the open issue opencharly/opencharly#431.
• R3 - nothing duplicated: the reference is changed once, in place.
• R4 - no workaround: no comment-out, no TODO, no alias.
• R5 - the sweep is executed and pasted above, showing the changed file's matching lines and the main baseline.
• R6 - user work preserved and Git safety kept: a feat/ branch off a just-fetched origin/main in a session-owned clone; no force-push, no hook bypass, no direct push to main.
• R7 - behaviour, not compilation: prose only; no runtime path is claimed to have run, and the retirement evidence is repository state read live.
• R8 - the artifacts at this boundary are the changed file and the retired repositories' own files, read at their source.
• R9 - the CalVer-stamped worktree binary was used for the surrounding cutover work.
• R10 - no behavioral field moved, so no runtime target is affected.

Change classification

Documentation — prose-only lines changed in a skill block. No behavioral field moved.

Tier: documentation reviewed.

Agent: openclaw-family
Assisted-by: DSH DeepSeek deepseek-v4.1-flash (documentation reviewed)

[github-actions]

Review — BLOCK

Head SHA: a8af412a3bfe2f150ea6a91a476256b615bf8753 (branch: feat/drop-retired-openclaw-refs-v3)

Change class: docs/skill prose inside charly.yml (1 file, +9/−10) — the diff is entirely inside the container-nesting-skill: text block.

Comment thread: 0 comments — nothing to disposition.

Blocks

1. R1 — the skill now diverges from reality, and the body's own evidence proves it.
   The body's "Liveness" section is the proof that the removed entities are still live:
   grep -c … = 3 (not 0) and gh issue view 431 → OPEN, with the removal carried by the
   unmerged opencharly/marketplace#434. So /charly-openclaw:openclaw-desktop still
   exists and still composes this candy, yet this PR removes it from the skill's
   Used In Boxes list (charly.yml, the @@ -827 hunk) and genericizes its example
   (@@ -700). R1 makes a skill↔reality divergence an incident. The body parks the gap on
   #431/#434, but that does not cure the divergence this tree now carries: the skill
   documents a post-retirement state that does not exist. Fix: land the retirement in the
   same phase as this skill edit (i.e. do not drop a live box's reference while it is live),
   or paste executed evidence that the box/repos are actually retired.

2. A1 — the body misstates the diff it claims to account for.
   The Summary asserts "The removed lines named entities the OpenClaw family cutover
   retires." That is false for the removed line charly-fedora: (@@ -716): charly-fedora
   is the live box /charly-distros:charly-fedora, which the body itself keeps in the
   added lines ("Evaluating the security posture of /charly-distros:charly-fedora, …"). A
   removed line named a live entity, contradicting the body's stated principle ("Where a line
   named a live entity it is kept"). Fix the Summary to account for the charly-fedora →
   my-root-box rename honestly.
   Secondary accounting gap: the body states 9 insertions but its "Added lines" list has
   only 8 entries — the blank line that replaces - openclaw-full in the example (the
   @@ -700 hunk) is unlisted.

3. R5/B13(d) — the pasted sweep does not key to the removed identifiers.
   The grep pattern opencharly/(layer-openclaw-full|layer-charly-openclaw|pod-openclaw)
   matches none of the identifiers actually removed (openclaw-desktop, openclaw-full,
   charly-fedora); it is a proxy for marketplace repo refs, not a self-test of the removed
   references. R5/B13(d) require actually-executed output keyed to the swept identifiers.
   Paste a command whose pattern matches the removed strings, with its real output.

Verdict: BLOCK

---

ollama-cloud/deepseek-v4.1-flash — action-review.

View action run

[github-actions]

Review — PASS

Head SHA: a8af412a3bfe2f150ea6a91a476256b615bf8753 (branch: feat/drop-retired-openclaw-refs-v3)

Security screen: PASS — prose-only change inside the container-nesting-skill: block; no secrets, no network/code, no guardrail weakening, no embedded directives to the reviewer.

Checklist: All applicable rules PASS or NA (docs/skill-prose class; no runtime/bed or ADE/SDD items apply).

Comments considered (1):

• github-actions[bot] BLOCK → stale-superseded. It was derived from an earlier revision whose body carried a "Liveness" section (grep -c … = 3, gh issue view 431 → OPEN). The CURRENT body instead supplies executed gh api retirement evidence (RETIRED tombstones + 0 entity declarations for layer-charly-openclaw/layer-openclaw-full/layer-openclaw-full-ml) and a sweep keyed to the removed identifiers (openclaw-(desktop|full|full-layer|full-ml-layer) → 9 on main, 0 on branch). That 9/0 count is internally consistent with the diff: exactly 9 removed lines match that pattern (all the openclaw-desktop/openclaw-full lines), and none are added. All three prior findings (R1 liveness, A1 charly-fedora misstatement, R5 pattern keying) fail to reproduce against the current diff/body. The charly-fedora → my-root-box example rename is an illustrative genericization; the live /charly-distros:charly-fedora is still correctly listed under "Used In Boxes", so no skill↔reality divergence results.

Verdict: PASS

---

ollama-cloud/deepseek-v4.1-flash — action-review.

View action run