fix: update checkpoint search commands

[Anexria]

Summary

Update the search skill and its Codex/Claude search agents for the grouped checkpoint commands shipped by Entire CLI 0.8.42.

The previous instructions used the removed entire search / entire explain forms. The updated workflow uses entire checkpoint search and entire checkpoint explain, defers uncertain flags to installed entire agent-help, and distinguishes an unavailable server index from genuinely empty local history.

Skill changed

• No skill changed
• New skill
• Existing skill: search

Validation

• npx -y skills@latest add . --list --yes
• JSON metadata validated, if changed — not applicable; no JSON changed
• OpenCode plugin syntax checked, if changed — not applicable; plugin shim unchanged
• Markdown previewed, if changed
• quick_validate.py skills/search reports Skill is valid!
• Fresh-context forward test recovered a known checkpoint after server search returned zero results

Example prompt

If skill behavior changed, include a prompt that demonstrates the new or updated behavior.

Find the earlier Entire session where we drafted the responsive-mode follow-up issues. If normal search has no hits, check whether local checkpoints still exist before concluding there is no history.

Notes

• entireio/cli#1841 reports the observed server-backed search behavior: known local checkpoint content can return zero results or HTTP 500.
• The fallback stays within Entire checkpoint commands and summarizes evidence instead of dumping transcripts.

[SparshM8]

Great catch on the search skill — I was hitting the same broken commands after upgrading to CLI 0.8.42.

While reviewing this PR I noticed the same deprecated entire search and entire explain commands appear in three other skill files that aren't covered here:

skills/recall/SKILL.md
The Guardrails section explicitly locks the agent to the old commands — "Use only the canonical Entire commands for this skill: entire search and entire explain." — and the Process section runs entire search "" --json directly. Because the Guardrails forbid alternatives, the agent has no fallback when the command fails.

skills/session-to-skill/SKILL.md
Rules section says to prefer entire search and entire explain, and the Workflow runs entire search "" --json. The error-handling guidance is also written around the old command's output format, so the fallback logic is incorrect on 0.8.42+.

skills/what-happened/SKILL.md
Rules section says to prefer entire explain --commit and explicitly tells the agent not to use entire why as an alternative. On 0.8.42+ this leaves the agent with a broken provenance lookup and no valid fallback path.

If it would help, I'm happy to open a follow-up PR that covers these three files with the same fix pattern used here — or I can add them to this PR if you'd prefer to land it all together. Happy to coordinate whichever way works best for the maintainers.