RULES.md: settle the writeup's contradictions by date

[asubbarao]

The what-alok-said writeup listed seven places where his messages disagree with RULES.md. RULES.md was curated on 10-17, so they are settled by date, the writeup's own rule (newer wins):

• Newer (10-18), RULES.md changes: 12 allows a plain concat for a statement with few holes (templates are for the four shapes); 5 allows one ->> per receipt body field until quackapi returns typed replies.
• Older, RULES.md stands: dispatch even one statement, dollar quotes, WITH ORDINALITY over array_agg ORDER BY, no regex on backend data, dispatch shape mandatory.

The writeup's section 6 now records the outcome instead of open questions. No hook changes: no check enforces concat chains, and sql_json_extract only warns.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NCnrCZ6kWmagfJjoK8jVGQ

[asubbarao]

duckstack/review: failure

• skills/self-dispatch/RULES.md:27 RULES.md 5 / check 7 (results come back as columns): The allowance names no shape, so receipt.body ->> 'x' repeated on every field across a query reads as allowed, which is the '100 line ->>' pattern he rejected. 'One per field' also doesn't say where the extraction should happen.
  Rewrite: Until quackapi returns typed replies, a receipt body field may be read with one ->> (no chained arrows), once, in the CTE right after the unnest, which names it as a column for every later reader; no ladders, no from_json structures (10-18).
• skills/what-alok-said/writeups/2026-10-18.md:393 RULES.md 25 (delete wrong advice; settle contradictions instead of piling them up): Section 6 now settles concat and the single ->>, but sections the PR left alone still contradict it: 2.2 says 'section 6 holds the open concat question' and that the ->> allowance 'is in section 6', 2.10 still says 'no multi-part concat (RULES.md 12, 17; see section 6)' and that the 10-14 ORDER BY message waits 'until he rules', and section 4 item 6 plus its closing line say two rules 'wait on his ruling'. Those are now wrong advice.
  Rewrite: In the same PR, update the stale lines. 2.2: 'Render each row's statement into a sql column (word arrays plus quote_literal, a named template, or a plain concat for few holes; RULES.md 12).' and 'One ->> per receipt body field is allowed until quackapi returns typed replies (RULES.md 5); ladders and from_json structures are out.' 2.2 ordering: 'use WITH ORDINALITY / array_sort (RULES.md 2); the older 10-14 array_agg(... ORDER BY ...) message lost (section 6).' 2.10: 'Otherwise statements are word arrays plus quote_literal, or a plain concat with few holes (RULES.md 12).' Section 4: delete item 6 and change the closing line to 'That is 40 rules a hook could catch, plus the 18 named at the end of section 3.'
• skills/self-dispatch/RULES.md:49 RULES.md 17 / check 6 (templates are few, canonical, named by shape): The added list says there are four template shapes (sql, shell, network, bash loop fallback), but rule 17 in the same file says templates/ holds five (shell, pipeline, sql, network, loop), so the file now disagrees with itself about which templates exist.
  Rewrite: 12. Readable strings. Arrays of words, a canonical template with holes (rule 17: shell, pipeline, sql, network, loop), or a plain concat when a statement has only a few holes (10-18); no || chains, no chr(39)/chr(10)

Opus 5.5, duckstack review queue (duckstack-review skill)

[asubbarao]

duckstack/review: failure

• skills/self-dispatch/RULES.md:50 RULES.md 25 (delete wrong advice) / duckstack-review check 6: Rule 12 now allows a plain concat, but this PR leaves duckstack-review check 6 ('Statements come from word arrays and quote_literal, or a generic tera template') and manifesto §3 unchanged. Reviewers will flag the newly allowed concat, and 'few holes' has no bound, while writeup section 3 calls 'concat chains over two parts' hook-checkable.
  Rewrite: 12. Readable strings. Arrays of words, a canonical template with holes (rule 17: shell, pipeline, sql, network, loop), or one plain concat of at most two holes (10-18); no || chains, ... — and in the same PR change duckstack-review SKILL.md check 6 to: "Statements come from word arrays and quote_literal, a generic tera template that only substitutes, or one concat of at most two holes (RULES.md 12)."
• skills/what-alok-said/writeups/2026-10-18.md:391 RULES.md 17 / RULES.md 25: This line says templates are for four shapes (sql, shell, network, bash loop fallback), but RULES.md 12 and 17 as changed in this PR list five canonical templates (shell, pipeline, sql, network, loop), so the settled outcome disagrees with the rule it claims to record.
  Rewrite: - Concat for a statement with few holes (10-18): templates are for the canonical shapes in RULES.md 17 (shell, pipeline, sql, network, loop); a plain concat is enough otherwise. RULES.md 12 now says so.

Opus 5.5, duckstack review queue (duckstack-review skill)

[asubbarao]

duckstack/rules: success

• existing sql_vague_cte_names in skills/duckstack-review/SKILL.md#2.sql: Name every CTE for what it holds (content_first_100, self_dispatch), not t / x / sd / tmp.
• existing sql_vague_cte_names in skills/duckstack-review/SKILL.md#5.sql: Name every CTE for what it holds (content_first_100, self_dispatch), not t / x / sd / tmp.

duckstack review queue, rule gate (hooks/hook_checks.sql)

[asubbarao]

duckstack/review: failure

• skills/what-alok-said/writeups/2026-10-18.md:232 RULES.md 12, 17: 'a plain concat with few holes' and 'Templates only for statements with many holes' (line 231) don't match RULES.md 12/17. Those rules allow any canonical-shape template and limit concat to one call with at most two holes.
  Rewrite: - Templates come from the canonical set by shape (RULES.md 17: shell, pipeline, sql, network, loop). Otherwise statements are word arrays plus quote_literal, or one plain concat of at most two holes (RULES.md 12).
• skills/what-alok-said/writeups/2026-10-18.md:61 RULES.md 12: 'a plain concat for few holes' is looser than RULES.md 12's 'one plain concat of at most two holes', so a reader of the writeup could justify a concat with three or more holes.
  Rewrite: (word arrays plus quote_literal, a named template, or one plain concat of at most two holes; RULES.md 12)
• skills/self-dispatch/RULES.md:27 RULES.md 13 (internal contradiction): The new allowance of one ->> per receipt body field clashes with rule 13, which still bans 'selector-taking extractors in committed code' with no exception, so the rules file now says two different things about the same expression.
  Rewrite: In rule 13 change the clause to: "json_extract ladders and selector-taking extractors in committed code (except the single ->> per receipt body field allowed by rule 5)", and keep line 27 as written.

Opus 5.5, duckstack review queue (duckstack-review skill)

[asubbarao]

duckstack/rules: success

• existing sql_vague_cte_names in skills/duckstack-review/SKILL.md#2.sql: Name every CTE for what it holds (content_first_100, self_dispatch), not t / x / sd / tmp.
• existing sql_vague_cte_names in skills/duckstack-review/SKILL.md#5.sql: Name every CTE for what it holds (content_first_100, self_dispatch), not t / x / sd / tmp.

duckstack review queue, rule gate (hooks/hook_checks.sql)

[asubbarao]

duckstack/review: success

No findings.

Opus 5.5, duckstack review queue (duckstack-review skill)