Generic tool results: render structured content items (#227)

[cboos]

Problem

A tool result for a tool without a specialized output renderer (e.g. ToolSearch) can carry a content list of typed items other than text/image — notably tool_reference blocks. The generic fallback formatter only extracted text and images, so such results rendered nothing on the result side (the tool_use showed, but the result was blank). Fixes #227.

Fix

• HTML (html/tool_formatters.py, format_tool_result_content_raw): the list-content loop now collects leftover typed items into structured_items, rendered by a new _structured_items_html helper through the existing JSON/Markdown hybrid table — the same machinery used for JSON results. This inherits the depth and per-container breadth caps for free (a 250-item list falls back to a <pre> JSON dump rather than generating a 250-row table). Wired into all tails: the images branch appends it; the no-images branch returns it (alone, or after text). String, text-only, image-only, and error-string paths are unchanged.
• Markdown (markdown/renderer.py, format_ToolResultContent): list content now renders text items as text and remaining typed items as a JSON block, instead of dumping the whole list verbatim. The TodoWrite special-case and the string path are unchanged.

The content array is the source (it carries the structured items directly), rather than toolUseResult.

is_error + structured items — deliberate choice

When a result is both is_error and carries typed content items, it renders the structured table rather than plain error text. This is intentional and verified safe: the is_error → read-as-text guard only applies to the JSON-string path, and genuine tool_reference (and other typed) items are not error message text. XSS posture is preserved on this path (structured items route through the same escape=True hybrid-table leaves; the breadth-cap fallback HTML-escapes too). Pinned by tests so a future refactor can't reorder the guard ahead of the structured path.

Tests

New test/test_generic_tool_result.py (13 cases): HTML + Markdown coverage, including the original bug (tool_reference-only → non-empty table), mixed text+structured, the is_error+structured pin, XSS escaping, breadth-cap fallback, and regressions on string / text-only / image-only / TodoWrite paths. Verified end-to-end on a real fixture (a ToolSearch result now emits structured tables where it previously rendered nothing). Full just ci green; no snapshot churn.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes
  • Structured tool results are now shown instead of being dropped.
  • Mixed text, structured data, and image outputs now render together more consistently.
  • Markdown output now separates plain text from structured JSON content for clearer display.
• Tests
  • Added coverage for structured tool-result rendering, including mixed content, error cases, and HTML escaping.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: a483a722-adc8-42e7-a2f8-3be2fa40e331

📥 Commits

Reviewing files that changed from the base of the PR and between 8013b61 and 99dd782.

📒 Files selected for processing (3)

• claude_code_log/html/tool_formatters.py
• claude_code_log/markdown/renderer.py
• test/test_generic_tool_result.py

---

📝 Walkthrough

Walkthrough

HTML and Markdown tool result renderers are updated to collect non-text/non-image structured content items (e.g., tool_reference) instead of silently dropping them. A new HTML helper converts these into the existing params-style JSON/HTML table, with a pre-block fallback. The Markdown renderer splits text and typed items into separate output sections. A new test module validates both renderers across all relevant cases.

Changes

Structured Tool Result Rendering

Layer / File(s)	Summary
HTML helper and format_tool_result_content_raw wiring claude_code_log/html/tool_formatters.py	Adds _render_structured_items helper (params-style table with JSON pre fallback); updates the item loop to collect non-text/non-image items into structured_items; threads structured_html into both the has-images and no-images output paths.
Markdown split rendering claude_code_log/markdown/renderer.py	Replaces the single JSON dump of the entire output.content list with separate rendering: type == "text" entries emit as stripped plain text, all other typed entries emit as a fenced JSON block, joined by blank lines.
Tests: HTML and Markdown structured content test/test_generic_tool_result.py	New test module with _result/_md_message helpers and _TOOL_REFERENCE fixture; TestHtmlStructuredContent covers non-dropped items, multi-reference, mixed text, HTML escaping, error parity, and legacy paths; TestMarkdownStructuredContent covers fenced JSON output, text separation, error parity, plain strings, and the TodoWrite special case.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• daaain/claude-code-log#216: Both PRs modify format_tool_result_content_raw in claude_code_log/html/tool_formatters.py to change how tool result content is rendered into the params-style hybrid JSON/HTML table.

Poem

🐇 Hop hop, what's this I see?
A tool_reference hiding from me!
No more dropped items, they'll all appear,
In tidy JSON tables, crystal clear.
Text splits from structured, each in its place —
The rabbit renders all with style and grace! ✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main change: rendering structured generic tool results.
Linked Issues check	✅ Passed	The PR addresses #227 by rendering structured content items for generic tools in both HTML and Markdown.
Out of Scope Changes check	✅ Passed	The changes stay focused on generic tool-result rendering and tests, with no obvious unrelated additions.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch dev/generic-tool-result

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[cboos]

@coderabbitai full review

[coderabbitai]

✅ Action performed

Full review finished.