docs(review-flows skill): daemon prerequisite, roles, and failure handling

[realtonyyoung]

What

Enhances the review-flows skill (kcap/skills/review-flows/SKILL.md) so the driving agent knows what has to be true for a flow to work, and what to do when it isn't.

• Roles — the driver calls these tools; the reviewer is a separate hosted Codex agent the server launches on a kcap daemon (never configured from the driver).
• Prerequisites — start_review_flow needs kcap login and a running daemon with this repo checked out, or it errors instead of returning a flow_run_id.
• Failure handling — on a no-daemon / ambiguous-daemon error, surface the fix (kcap daemon start -d with the target repo) rather than retrying blindly; don't start a flow (which spends real reviewer compute) unprompted.

Why

Setting the feature up end-to-end surfaced that these prerequisites were undocumented on the agent-facing surface — the skill explained the loop mechanics but not the daemon dependency or the driver/reviewer split, so an agent hitting a no-daemon error had no guidance.

Scope / relationship to other work

Doc-only; no CLI surface change (no new command/flag), so no README sync is required by the repo convention.

Complements #217 (AI-1056), which auto-registers the flows MCP server for Claude Code and rewrites the README/registration guidance. This PR touches only SKILL.md (which #217 does not), so the two don't conflict.

Linear: AI-774 (feature), AI-1056 (auto-register follow-up).

[qodo-code-review]

Qodo reviews are paused for this user.

Troubleshooting steps vary by plan Learn more →

On a Teams plan?
Reviews resume once this user has a paid seat and their Git account is linked in Qodo.
Link Git account →

Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center?
These require an Enterprise plan - Contact us
Contact us →

[realtonyyoung]

Found one documentation correctness issue in the new failure-handling guidance. The rest of the added sections look consistent with the flow tool surface, the daemon launch path, and PR #217 auto-registration wording.

[realtonyyoung]

Fixed in 1039c4d. The failure-handling guidance now splits the two cases instead of prescribing "start a daemon + retry" for both:

• No daemon → start one (kcap daemon start -d) and retry.
• Ambiguous match ("multiple daemons/checkouts match this repo") → retrying won't help; the tool exposes no daemon_name/repo_path, so the agent can't disambiguate. The skill now says to surface it to the user rather than loop, and references AI-1112 (the tracked fix: server-side auto-select by the requester's repo root).

Docs-only; no CLI surface change.

[realtonyyoung]

Rebased onto current main to clear the conflict (main moved a lot: #238 opt-in gating, #235 hosted-reviewer misfire guard, and #233/#234 vendor-uniform hosting all rewrote SKILL.md).

Given those, I trimmed this PR rather than force my original text back in:

• Dropped the "Roles" section — now redundant with main's intro + "When NOT to use" gating.
• Dropped the Codex-specific wording — post-[AI-1124] daemon: collapse hardcoded vendor allowlist into the launcher lookup #234 the reviewer's vendor/model come from the flow definition, not hardcoded Codex.

What remains is the still-unique, vendor-neutral bit main doesn't cover: a Prerequisites section (start_review_flow needs kcap login + a daemon with this repo checked out) and split failure handling — no-daemon (start one, retry) vs ambiguous multi-checkout match (can't disambiguate; surface it, AI-1112).

Now MERGEABLE (+12/−0, one file). Force-pushed; earlier resolved threads were anchored to the pre-rebase text.