[refactor] idd-list PR body parsing: ignore #N inside fenced code blocks (sister concern from #13)

[kiki830621]

Problem

idd-list v2.51.0+ 將實作 PR body parsing 偵測 #N references。第一版接受 false positive(#13 R1 risk acknowledged):fenced code block 內的 #N 也會被當成 issue ref。

例:PR body 含

# Example: how to reference an issue
git commit -m "Refs #99"

#99 出現在 code block 內,不是真正的 ref,但會被 #([0-9]+)� regex 命中。

Type

refactor / observability quality

Why P3

• 不影響 happy path:大多數 PR body 不會在 code block 內列 issue refs
• false positive 結果:idd-list 顯示「issue #99 has open PR #N」是錯的,但 user 一查就知道
• 修起來簡單但邊界 case 多(nested fences、indented blocks、tilde fences),完整修需測試

Recommendation

實作 markdown-aware parser:strip fenced code blocks(\``...```+~~~...~~~` + 4-space indent blocks)before regex scan。

Source

surfaced during /idd-diagnose #13 sister concern surfacing (Step 3.6) — referenced as risk R1

Current Status

Phase: closed
Last updated: 2026-05-07 by idd-close

Resolution

• PR feat(idd-list): markdown-aware fence strip + cluster_leader config (cluster #14 #15) #43 merged → main
• cluster with [enhancement] idd-list cluster_leader config (lowest | primary) (sister concern from #13) #15 — markdown-aware fence strip in idd-list Step 3.5

[kiki830621]

Closing Summary (cluster #14 with #15)

Problem

idd-list v2.51.0+ Step 3.5 PR body parsing 用 regex #(\d{1,7})\b 在 PR body 中找 issue refs。第一版接受 false positive(#13 R1 acknowledged):fenced code block 內的 #N 也會被當成 issue ref。

例:PR body 含

\`\`\`bash
git commit -m "Refs #99"
\`\`\`

該 #99 不該被當 issue ref 計入 cluster detection,但會被。

Root Cause

v2.51.0 spec 設計時為了「不 over-engineer for v1」,接受 markdown-aware filter 為 follow-up issue。本 issue 是該 follow-up。

Solution

Cluster-PR with #15:

#14 part: Step 3.5 加 strip_fenced_code(body) helper 在 regex scan 前剝除 ... blocks(支援 ```bash / ```python 等 language tag 變體):

FENCE_RE = re.compile(r'```[\s\S]*?```', re.MULTILINE)

def strip_fenced_code(body):
    return FENCE_RE.sub('', body or '')

# Used before regex scan:
body = strip_fenced_code(pr['body'])
raw_refs = set(int(m) for m in PR_REF_RE.findall(body))

Trade-off documented(per #14 acknowledged scope):

• Inline code (`#99`) 不剝除 — false positive 罕見 + 剝除會誤傷合法 inline closes #99
• Indented code blocks (4-space prefix) 不剝除 — 在 PR body 中極為少見,detection cost > benefit
• Tilde fences (~~~ ... ~~~) 不剝除 — GitHub-flavored markdown 預設用 backtick fence

Verification

Lightweight self-verify PASS:

• Pseudocode 完整 + Python imports 一致
• 對舊行為 backward compat:無 fenced block 的 PR body 行為與 v2.51.0 完全相同
• Spec language 在 SKILL.md line 155 標明 acknowledged limits + line 171 explicit document trade-off

Changes

• 8675b82 feat(idd-list): markdown-aware fence strip + cluster_leader config (Refs [refactor] idd-list PR body parsing: ignore #N inside fenced code blocks (sister concern from #13) #14 [enhancement] idd-list cluster_leader config (lowest | primary) (sister concern from #13) #15)
• e366d53 (squash on main) cluster PR feat(idd-list): markdown-aware fence strip + cluster_leader config (cluster #14 #15) #43 merged

Related (cluster)

• [enhancement] idd-list cluster_leader config (lowest | primary) (sister concern from #13) #15 (paired — cluster_leader config refactor in same SKILL.md section)
• idd-list 應該顯示引用 issue 的 open PR 資訊 #13 (great-parent — primary feature, R1 acknowledged this v1 limitation)