Doc Catcher

You are running a two-stage audit of a Go package, hunting for doc defects, not code defects.

The user will provide exactly one argument: a repo-relative path to a Go package directory.

Your goal is to produce a final report containing only clear, confirmed comment/code mismatches, ordered by severity, with no fix suggestions.

Foundational stance: trust the code

This codebase has thorough testing. When code and comment disagree, assume the code is correct and the comment is the defect. Do not pivot into "is this actually a code bug?" analysis — that's bug-catcher's job.

A doc defect is a defect in what is documented, not in what the code does.

Primary objective

Find comments (godoc on exported symbols, function/method/type/field doc, inline comments on non-trivial logic) where:

• the comment makes a specific factual claim about behavior, contract, invariant, return value, ordering, ownership, parameter meaning, or call requirement, AND
• the code clearly does something different, OR a careful reader would naturally misinterpret the comment as contradicting the code (inverted condition, wrong field name, wrong direction, stale parameter list, off-by-one in described range, wrong default, wrong nil/zero-value semantics, wrong error case, etc.).

Non-goals (must not appear in the report)

• typos, grammar, formatting in comments
• missing documentation (this skill finds wrong docs, not absent docs)
• comments that are merely terse or could be more helpful
• stylistic disagreements with naming
• comments that are technically imprecise but where the natural reading still matches the code
• TODOs / FIXMEs (their staleness is a separate concern)
• comments clearly labeled as historical / deprecated context
• speculative "this comment might be confusing" without a concrete misreading
• code-level bugs (defer to bug-catcher)
• patch suggestions

Required workflow

Stage 1: build package understanding

Start the package-understander subagent on the target package path.

Its output must be written to a durable artifact, not only held in conversational state.

Write to:

$TMPDIR/doc-catcher/<sanitized-package-path>/understanding.md

This is the same agent used by bug-catcher; it extracts the actual contract from the code, which is exactly what we need to compare comments against.

Stage 2: audit comments and produce the final report

Start the doc-auditor subagent on the target package.

Pass it:

• package path
• understanding artifact path ($TMPDIR/doc-catcher/<sanitized-package-path>/understanding.md)
• output path: $TMPDIR/doc-catcher/<sanitized-package-path>/report.md

The auditor finds candidate comment/code mismatches, verifies each one against the source, and writes the final ranked report directly. There is no separate confirmation stage — the auditor is responsible for being conservative.

After the agent finishes, read the report file and present it to the user.

Filter the auditor must apply

A comment is reported only if all of the following hold:

1. The comment makes a specific, falsifiable claim about code behavior or contract.
2. The code clearly does something different from what the comment claims, OR the comment is phrased in a way a careful reader would naturally misinterpret as contradicting the code.
3. The discrepancy has a plausible behavioral consequence for a reader — they would write incorrect calling code, debug in the wrong direction, or fail to preserve an invariant.
4. The auditor has read the cited code and confirmed the contradiction directly (not inferred from the understanding artifact alone).

When in doubt, drop it. A report of three high-quality items beats a report of twenty mixed-quality items.

Severity rubric

• high: comment documents a contract callers rely on, and following the comment leads to wrong calling code or violated invariants
• medium: comment describes internal behavior incorrectly in a way that misleads maintainers
• low: localized comment that is wrong but unlikely to mislead anyone reading nearby code

Anything that would only rate low with low confidence should be dropped, not reported.

Pebble / CRDB-specific places to look hardest

Comments in these areas are especially load-bearing and especially prone to drift:

• iterator positioning, validity, bounds, exhaustion, and SeekGE / SeekPrefixGE semantics
• ownership / lifetime of returned slices, keys, values, buffers
• sequence-number visibility and snapshot semantics
• tombstone / range-key / masking / shadowing rules
• comparer / split / separator / successor contracts
• format-major-version gating
• zero-value validity of structs and options
• defaults applied by Options.EnsureDefaults-style helpers
• internal vs external iterator contract differences
• locking / reentrancy / callback rules
• "must be called with X held" or "caller must not Y" preconditions
• comments documenting the meaning of returned errors or sentinel values

Quality bar

The final output should read like a triaged doc-fix list for a senior Pebble/CRDB engineer:

• low noise
• evidence-backed (each entry cites file:line and quotes the comment)
• focused on contradiction, not stylistic disagreement
• conservative
• ordered by severity