fix(report): relay the report summary as chat text, not a code block

[theCodeDrift]

At the end of onboarding, agents consistently pasted the taskless report summary into a fenced code block. Inside the fence the tskl.es link isn't clickable and the rules don't render as a list, which makes the report the whole session built toward hard to follow.

The report recipe asked for this. Step 4 said to show the summary "exactly as printed", and agents read that as verbatim, which in chat means a code fence.

Step 4 now says:

• Relay the summary as ordinary chat text, not inside a code block or quote, and says why.
• Keep the summary's content and order; do not add, drop, or reword facts. The CLI template still decides what is said.
• Put the link on its own line as a plain URL or a Markdown link, and render the rules as a list.
• Drop the extra "it can be deleted" line the agent was told to add, because the CLI summary already ends with it.

The topic goes from v3 to v4, and test/report.test.ts pins the new version.

No changeset: taskless report is not in a release yet, and the pending streamlined-onboarding-report changeset covers it.

[theCodeDrift]

@claude /review

[github-actions]

Claude finished @theCodeDrift's task in 1m 23s —— View job

---

Review mode: incremental — read 0 prior review thread(s) before reviewing.

.prior-review.json contains no review threads, no review summaries, and no top-level comments other than the triggering @claude /review itself. This is the first review of this PR.

Summary

The change is small and does exactly what the PR description claims:

• report.md step 4 is reworded to tell the agent to relay the CLI's report summary as ordinary chat text (not a fenced block/quote), explains why (unclickable link, buries the report), asks for content/order to be preserved, calls out the link-on-its-own-line + rules-as-a-list formatting, and drops the redundant "it can be deleted" instruction.
• Topic version bumped v3 → v4, and report.test.ts's regex is updated to match.

I checked this against the actual renderer, packages/cli/src/report/summary.ts:

• The rules are already emitted as - rule.title (...) lines (a Markdown list), consistent with "render the rules as a list."
• The link line is already View it: ${url} on its own line, consistent with "link on its own line."
• The summary's last line is already It can be deleted after review. Ask, or run ..., which is exactly the line the recipe no longer asks the agent to add redundantly.

So the new instructions accurately describe the real CLI output rather than asserting something that isn't true. I also confirmed no other agent recipe file still has the old "exactly as printed" wording, and no other string in the repo still references "topic v3" for the report topic, so the version bump is self-contained.

I found no correctness, security, or clarity issues in this diff. One minor, non-blocking observation:

• [New] report.test.ts's updated test only asserts the version-string change; it doesn't assert anything about the new "relay as chat text / not fenced / render as a list" instructions actually appearing in the rendered topic text (the way the test already checks other specific phrases elsewhere in the file, e.g. "**Only review comments count.**"). Given this is a prompt/recipe text file rather than executable logic, and the PR is adding prose guidance rather than new behavior, I don't consider this a blocking gap — just worth knowing if a future edit to step 4 silently drops one of the new clauses, there's nothing pinning it down besides the version bump.

The no-changeset rationale checks out: taskless report is introduced by the still-pending .changeset/streamlined-onboarding-report.md, so there's no released surface this PR needs its own release note for.

[theCodeDrift]

Re: @github-actions — "Claude finished @theCodeDrift's task in 1m 23s… report.test.ts's updated test only asserts the version-string change…"
#523 (comment)

Added in 6dc8c77: a test that pins the new step 4. It asserts the relay-as-chat-text and link-on-its-own-line instructions are present, that "exactly as printed" is gone, and that the agent is told not to repeat the delete line.

— AI Coding Agent