Fix version conflict reporting in summarize-changes skill (#1087)

* Fix version conflict reporting in summarize-changes skill

- Add fallback for fetching toml from merged PRs (head branch -> merge commit -> main)
- Add critical instruction to faithfully report script output, not fabricate results
- Clarify version conflict definition: same type + same ext + same range required
- Add Subdir column to conflict table for clarity

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Use PR-specific temp file paths to avoid parallel collisions

When multiple summarize-changes analyses run concurrently, they all
wrote to /tmp/pr_diff.txt and /tmp/pr_toml.txt, causing data races.
Use /tmp/pr_<number>_diff.txt and /tmp/pr_<number>_toml.txt instead.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: wmitsuda

.claude/skills/summarize-changes/SKILL.md

@@ -31,27 +31,35 @@ Do NOT proceed with any further steps.
 
 ## Step 2: Fetch PR Data and Run Analysis
 
-Run these commands using the Bash tool:
+Run these commands using the Bash tool. **IMPORTANT:** Use the PR number in temp file paths to avoid collisions when multiple analyses run in parallel.
 
 1. `gh pr view <number> --repo erigontech/erigon-snapshot` to get PR title, description, and metadata
 2. Save the diff to a temp file:
    ```
-   gh pr diff <number> --repo erigontech/erigon-snapshot > /tmp/pr_diff.txt
+   gh pr diff <number> --repo erigontech/erigon-snapshot > /tmp/pr_<number>_diff.txt
    ```
-3. Fetch the full toml file from the PR's head branch:
+3. Fetch the full toml file from the PR's head branch (with fallback for merged PRs where the branch may be deleted):
    ```
    HEAD_REF=$(gh pr view <number> --repo erigontech/erigon-snapshot --json headRefName -q .headRefName)
+   MERGE_COMMIT=$(gh pr view <number> --repo erigontech/erigon-snapshot --json mergeCommit -q .mergeCommit.oid)
    TOML_FILE=$(gh pr diff <number> --repo erigontech/erigon-snapshot --name-only | head -1)
+   # Try head branch first; if deleted (merged PR), fall back to merge commit, then main
    gh api "repos/erigontech/erigon-snapshot/contents/$TOML_FILE" \
-     -H "Accept: application/vnd.github.raw" -F ref="$HEAD_REF" > /tmp/pr_toml.txt
+     -H "Accept: application/vnd.github.raw" -F ref="$HEAD_REF" > /tmp/pr_<number>_toml.txt 2>/dev/null \
+   || gh api "repos/erigontech/erigon-snapshot/contents/$TOML_FILE" \
+     -H "Accept: application/vnd.github.raw" -F ref="$MERGE_COMMIT" > /tmp/pr_<number>_toml.txt 2>/dev/null \
+   || gh api "repos/erigontech/erigon-snapshot/contents/$TOML_FILE" \
+     -H "Accept: application/vnd.github.raw" -F ref="main" > /tmp/pr_<number>_toml.txt
    ```
 4. Run the analysis script with both files:
    ```
-   python3 "$(git rev-parse --show-toplevel)/.claude/skills/summarize-changes/analyze_diff.py" /tmp/pr_diff.txt /tmp/pr_toml.txt
+   python3 "$(git rev-parse --show-toplevel)/.claude/skills/summarize-changes/analyze_diff.py" /tmp/pr_<number>_diff.txt /tmp/pr_<number>_toml.txt
    ```
 
 The script (`analyze_diff.py` in the skill directory) parses the diff, classifies all changes, and outputs structured sections. Use its output to build the final report.
 
+**CRITICAL: You MUST faithfully report the script's output.** Do NOT fabricate, guess, or paraphrase results. For every critical section (hash changes, unexpected deletions, version conflicts, version downgrades), copy the exact counts, file types, and ranges from the script output. If the script says `count=0` for version conflicts, report zero — do not infer conflicts from other observations.
+
 ### What the script detects
 
 - **Hash Changes**: same filename in both removed and added sets with different hash (CRITICAL)
@@ -127,19 +135,25 @@ If NO unexpected deletions, use ✅ emoji:
 
 ### VERSION CONFLICTS
 
-If version conflicts exist, use 🚨 emoji:
+A version conflict means the SAME file type, SAME extension, AND SAME range (identical start-end) has multiple versions in the final toml. For example, both `accessor/v2.0-logaddrs.0-64.efi` and `accessor/v2.1-logaddrs.0-64.efi` existing simultaneously is a conflict. Different versions covering DIFFERENT ranges (e.g., v2.0 for 0-256 and v2.1 for 256-512) is NOT a conflict — that is a normal version transition boundary.
+
+**CRITICAL: Only report conflicts that appear in the script output.** The script already applies the correct definition (same type + same range). Copy the exact types, ranges, and counts from the `=== VERSION CONFLICTS ===` section. Do NOT infer or fabricate conflicts.
+
+If version conflicts exist (script reports count > 0), use 🚨 emoji:
 
 ### 🚨🚨🚨 VERSION CONFLICTS — DO NOT MERGE
 
 > **Multiple versions of the same file for the same range must not coexist. This will cause download conflicts for nodes.**
 
-Group by State Snapshots / CL Snapshots / EL Block Snapshots. Show a table:
+Group by State Snapshots / CL Snapshots / EL Block Snapshots. Show a table with one row per conflict from the script output:
+
+| Subdir | Type | Ext | Range | Versions |
+|--------|------|-----|-------|----------|
+| accessor | accounts | .vi | 0-32 | v1.1, v1.2 |
 
-| Type | Ext | Range | Versions |
-|------|-----|-------|----------|
-| accounts | .vi | 0-32 | v1.1, v1.2 |
+When many conflicts share the same pattern (same subdir, type, ext, versions) across consecutive ranges, they can be summarized as a single row with a range span, e.g., "0-64 through 1792-1856 (29 ranges)".
 
-If NO version conflicts, use ✅ emoji:
+If NO version conflicts (script reports count=0), use ✅ emoji:
 
 ### ✅ Version Conflicts