feat: --body-regex filter + --help-llm compact cheatsheet

[brunojm]

Summary

Third PR in the LLM-friendly series. Closes out the core feature set from the improvement list: filter/help surface area for agents.

What changed

`--body-regex REGEX`

Regex variant of `--body-grep`, mirroring how `--url` pairs with `--url-regex`. Compiled at CLI parse time so bad patterns error with exit code 2 before any file is read. Use `(?i)pattern` for case-insensitive matching. Composes as AND with `--body-grep` and all other filters.

```bash
hargrep --body-regex '(?i)session\sexpired' recording.har
hargrep --body-regex '"error":\s".*timeout"' --status-range 5xx recording.har
```

`--help-llm`

Prints a compact flag reference and exits. 1566 bytes vs 3511 for clap's default `--help` — 55% smaller. Tuned for LLM consumption: one line per flag group, no examples, exit codes documented.

```
$ hargrep --help-llm
hargrep — HAR query CLI. Reads FILE (or stdin).

FILTERS (AND-combined):
--method GET|POST|... --status CODE
--status-range 4xx|200-299 --url SUBSTR
--url-regex REGEX --header 'NAME[:VALUE]'
--mime SUBSTR --min-time MS
--body-grep SUBSTR --body-regex REGEX

OUTPUT (mutually exclusive):
(default) Filtered entries as JSON (pretty in TTY, compact when piped).
--output json|jsonl|summary
--fields F,F,... id,url,method,status,status-text,time,mime-type,started-date-time
--count Matching entry count.
--overview {entries,status,methods,mime_types,top_domains,total_body_size_bytes,total_time_ms}
--domains [{domain,count}] sorted by count desc.
--size-by-type [{mime_type,total_bytes,count}] sorted by total_bytes desc.
--redirects [{id,url,status,location}] for every 3xx.
--entry N One entry by id (original 0-indexed HAR position).
...
```

This lets an agent discover flags on demand for ~400 tokens instead of carrying a 1k+ token cheatsheet in every system prompt.

Test plan

• `cargo test` — 129 tests pass (55 unit + 74 integration). New tests cover:
  • `--body-regex` matching response vs request bodies.
  • Invalid regex errors at parse time with exit code 2.
  • `--body-regex` composes with `--body-grep` as AND.
  • `--help-llm` output size < 2KB and contains every top-level flag.
  • `--help-llm` works without a file argument.
• `cargo clippy --all-targets -- -D warnings` — clean.
• `cargo fmt --check` — clean.

Follow-up (no further PRs planned in this series)

With PR 1, PR 3 (née 2), and this one merged, hargrep has the feature set the eval identified as missing. Natural next step: re-run the step-1 eval against the current `main` to measure whether we've actually closed the gap to `jq` on small HARs and opened a lead on medium/large ones.

🤖 Generated with Claude Code

[brunojm]

Also addressed Codex's feedback on #3 in 7b90748: --entry now conflicts with --body-grep AND --body-regex (the earlier --entry conflict sweep missed --body-grep because it landed in a different PR, and --body-regex is new in this one). Existing conflict test extended to cover both.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: feb600e2ea

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Declare --body-regex incompatible with --entry

The new --body-regex option is defined as a filter, but it can still be passed with --entry; in run, the --entry path returns before any filters are applied, so the regex is silently ignored (for example, --entry 0 --body-regex '^nope$' still returns entry 0). This creates surprising behavior for the new flag and should be prevented by adding a conflict with --entry at argument definition time.

Useful? React with 👍 / 👎.