cwilliams5/review-static-history

Enter planning mode. Mine git history for bugs that shipped and were later fixed, to find patterns a static analysis check could have prevented. This is evidence-driven — only propose checks backed by real bug occurrences.

Methodology

Step 1 — Gather evidence

Review the last 50 commits (or more if the repo is active) for fix commits. Look for:

• Commit messages containing "fix", "bug", "crash", "revert", "missing", "forgot", "wrong"
• Diffs that add a global declaration, add Critical "Off", fix a DllCall signature, correct a variable name, etc.

For each fix commit, categorize the root cause:

• What class of bug was it? (missing declaration, wrong type, unguarded state, resource leak, etc.)
• Could a regex or source parser have detected it before the commit?
• Is the pattern already caught by an existing check?

Step 2 — Tally patterns

Group fix commits by root cause class. Use occurrence count as a signal, not a hard gate:

• 3+ occurrences → strong signal, propose with confidence
• 1–2 occurrences → still valid if the check is cheap (fits in a batch, simple regex) AND the bug was painful (silent corruption, shipped to users, caused a long debugging session)
• 0 occurrences but obvious risk → note as speculative, lower priority

Step 3 — Filter against existing checks

Cross-reference each pattern against the current pre-gate inventory:

Already caught (validate these are working):

• Missing global declaration → check_batch_simple_b.ps1 (switch_global sub-check) + query_global_ownership.ps1 -Check
• Cross-file private function calls → query_function_visibility.ps1 -Check
• Ownership violations → query_global_ownership.ps1 -Check

If a pattern IS already caught but the bug still shipped, that's a different problem — the check may have a gap or the developer bypassed pre-gate. Flag it.

Step 4 — Propose checks

For surviving patterns (not already caught, statically detectable):

Feasibility test for each:

• Write the detection regex/logic sketch
• Run it mentally against the current codebase — what would it flag?
• Estimate false positive rate (>20% false positives = not worth it)
• Estimate maintenance cost (will this break when code is refactored?)
• For low-occurrence patterns: is the check cheap enough (batch sub-check, simple regex) to justify the low frequency?

Hard limit: propose at most 2 new checkers. Keep the plan focused on the highest-value patterns.

Placement Guidance

For each proposed check, specify placement:

• Existing batch — if short and fits theme (e.g., a simple pattern check → check_batch_simple.ps1)
• New standalone — if complex or needs specialized parsing
• See review-static-speed for batch architecture context if needed

Plan Format

Section 1 — Evidence table (all fix commits analyzed):

| Commit | Message | Root Cause Class | Already Caught? | |--------|---------|-----------------|----------------| | abc1234 | Fix missing global in foo | Missing declaration | Yes — check_batch_simple | | def5678 | Fix Critical not released on early return | Unmatched Critical | No |

Section 2 — Pattern frequency:

| Pattern | Occurrences | Commits | Already Caught? | Severity | |---------|-------------|---------|----------------|----------| | Unmatched Critical "Off" | 4 | abc, def, ghi, jkl | No | Silent corruption |

Section 3 — Proposed checks (max 2, highest-value patterns):

| Pattern | Detection Sketch | False Positive Est. | Placement | Evidence | |---------|-----------------|--------------------|-----------|---------| | Unmatched Critical | Track Critical "On" → scan for return/continue without prior Critical "Off" | Low — structural | check_batch_guards.ps1 | 4 bugs in last 50 commits |

Ignore any existing plans — create a fresh one.