Fold Skill body into its tool_use block

[cboos]

Closes #93.

Claude Code's Skill invocation lands on disk as three disjoint entries:

1. assistant Skill tool_use
2. user tool_result with the literal string "Launching skill: <name>"
3. user isMeta=True entry whose sourceToolUseID matches (1) and whose text is the expanded skill body (markdown, often 100+ lines)

Rendered as-is, (3) appears as a "🧑 User (slash command)" block visually disjoint from (1) — which is exactly what #93 flags: "seeing the skill content as a user message (right aligned) breaks the 'flow'."

The linkage is already in the data. (3) carries a top-level sourceToolUseID equal to (1)'s tool_use id. No heuristic, no parent-chain walk, no proximity guess — a field lookup.

Changes

• UserTranscriptEntry.sourceToolUseID: Optional[str] — Pydantic field.
• MessageMeta.source_tool_use_id — propagated via meta_factory (getattr-with-default handles older transcripts).
• ToolUseMessage.skill_body: Optional[str] — set only when paired.
• New _pair_skill_tool_uses(ctx) in renderer.py: single-pass indexing of slash-command TemplateMessages by source_tool_use_id, then folds the body into each Skill ToolUseMessage and drops the slash-command + the redundant "Launching skill" tool_result. Runs right after _render_messages (before detail filtering) and reuses _reindex_filtered_context for the drops.
• HTML renderer overrides format_ToolUseMessage: appends the body via render_markdown_collapsible (same collapsible primitive as slash-command / compacted-summary).
• Markdown renderer does the same with raw passthrough.
• New .skill-body CSS — left border + inset padding — makes the folded body read as nested content under the params table.

Detail-level interaction

Pairing runs before _filter_template_by_detail. At HIGH the body survives alongside the tool_use — intended, since once paired the body is content of the Skill tool_use, not an independent message. MINIMAL/LOW continue to drop the entire Skill invocation because ToolUseMessage itself is in those exclude chains.

Tests (8 new in test/test_skill_pairing.py)

Template-level:

• Body folds into ToolUseMessage.skill_body.
• Slash-command TemplateMessage is consumed.
• "Launching skill: X" tool_result is dropped.
• Non-Skill tool_uses (e.g. Bash) untouched — their tool_result stays.
• isMeta entries without sourceToolUseID still render as standalone slash-commands.
• Orphan body (sourceToolUseID pointing at missing tool_use) stays standalone.

Renderer output:

• HTML: .skill-body class + rendered markdown present, "Launching skill" absent.
• Markdown: raw markdown passthrough, "Launching skill" absent.

Validation

• just test (excluding tui/browser/pagination): 922 passed, 7 skipped.
• uv run pyright on modified files: 0 errors, 0 warnings.
• uv run ty check: All checks passed (0 diagnostics).
• ruff format + ruff check: clean.
• Edge-case snapshot regenerated (CSS addition only, no behaviour change).

Review

Monk approved at dc1770a. Optional non-blocker flagged: the inner tool_result search in _pair_skill_tool_uses is O(N·M) for typical transcripts with a handful of Skills; could be tightened to O(N+M) by indexing tool_results once. Left for a later perf pass since the actual cost is negligible.

Summary by CodeRabbit

• New Features

  • Skill tool invocations are paired with their expanded slash-command bodies and rendered as a single block; HTML adds a collapsible presentation of the skill body and Markdown appends the body beneath tool params.
  • Pairing metadata is propagated so renderers can fold related user meta entries into tool-use renderings.

• Bug Fixes

  • Suppresses redundant "Launching skill" messages for paired Skill invocations.

• Style

  • Added styling to visually nest and separate skill body content.

• Tests

  • End-to-end tests verify pairing, scoping, suppression, and rendered output.

[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: ff871784-f078-49a9-825c-6704794600d9

📥 Commits

Reviewing files that changed from the base of the PR and between afb8e83 and d2baaac.

📒 Files selected for processing (1)

• claude_code_log/renderer.py

---

📝 Walkthrough

Walkthrough

Fold Skill slash-command user entries into their corresponding Skill ToolUseMessage by pairing on (session_id, tool_use_id), attach the skill markdown as skill_body, remove the original slash-command and matching non-error "Launching skill:" tool_result, and render the body inline (HTML collapsible / Markdown raw).

Changes

Cohort / File(s)	Summary
Data Model claude_code_log/models.py	Added optional linkage fields: UserTranscriptEntry.sourceToolUseID, MessageMeta.source_tool_use_id, and ToolUseMessage.skill_body.
Metadata Factory claude_code_log/factories/meta_factory.py	create_meta() now sets MessageMeta.source_tool_use_id from transcript.sourceToolUseID (or None).
Pairing Preprocess claude_code_log/renderer.py	New pass folds Skill ToolUseMessage with matching isMeta=True user slash-command by (session_id, source_tool_use_id): attaches skill_body, removes the matched slash-command and the canonical non-error "Launching skill:" ToolResultMessage, and reindexes ctx. Includes robust _is_launching_skill_payload detection.
Rendering claude_code_log/html/renderer.py, claude_code_log/markdown/renderer.py	Overrides format_ToolUseMessage() to append skill_body when present: HTML renders it as a collapsible markdown block (render_markdown_collapsible) with fixed section id "skill-body", Markdown appends the raw skill_body beneath params.
Styling claude_code_log/html/templates/components/message_styles.css	Added .skill-body CSS rules (top margin, padding, left border using --tool-param-sep-color, smaller font).
Tests & Snapshots test/test_skill_pairing.py, test/__snapshots__/test_snapshot_html.ambr	New end-to-end tests for pairing, scoping, edge cases, and guarded drop logic; updated HTML snapshot includes .skill-body CSS.

Sequence Diagram

sequenceDiagram
    participant Transcript as Transcript
    participant Pairer as SkillPairingLogic
    participant Messages as MessageStream
    participant Renderer as Renderer
    participant Output as RenderedOutput

    Transcript->>Pairer: Emit entries (tool_use "Skill", tool_result "Launching skill:", user meta with sourceToolUseID)
    Pairer->>Pairer: Detect Skill tool_use and find matching meta by (session_id, sourceToolUseID)
    Pairer->>Pairer: Extract slash-command body -> attach to ToolUseMessage.skill_body
    Pairer->>Pairer: Mark matched user/meta and canonical tool_result as consumed
    Pairer->>Messages: Remove consumed entries and reindex stream
    Messages->>Renderer: Provide ToolUseMessage (with skill_body)
    Renderer->>Output: Render tool_use + collapsible / appended skill body

Loading

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~45 minutes

Possibly related PRs

• Polish User Messages #60: Similar pairing and collapsible rendering changes touching pairing logic and renderer behavior.
• More Collapsible Content & Slash Command Support #50: Related renderer helpers (render_markdown_collapsible) and earlier isMeta/source-tool handling reused here.
• Introduce visual structure for the conversation and some specialized tool rendering #37: Prior renderer/CSS adjustments overlapping with the Skill folding and styling.

Poem

🐰 I found a Skill and gave a hop,
I tucked its body under the tool-use top.
No stray slash-command left to roam,
Skill and tool now cozy at home.
Hooray — a tidy render, off I hop!

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 60.71% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title directly describes the main change: folding Skill body content into the tool_use block, which matches the core objective of pairing skill invocations.
Linked Issues check	✅ Passed	The PR comprehensively implements issue #93 requirements: it embeds skill content into the Skill tool block via skill_body field, removes redundant user messages, and updates renderers to display the body nested within the tool_use block.
Out of Scope Changes check	✅ Passed	All changes are directly scoped to implementing the skill body pairing feature: model fields, renderer logic, CSS styling, factory updates, and comprehensive tests. No unrelated modifications are present.

_{✏️ 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/pair-skill-user-message

---

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 and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@claude_code_log/renderer.py`:
- Around line 2032-2091: The pairing currently matches only on tool_use_id and
can collide across sessions and can drop real results; modify
_pair_skill_tool_uses to key slash_by_source by (source_tool_use_id,
session_key) where session_key is taken from a session/render identifier on
messages (e.g. msg.meta.render_session_id or msg.meta.session_id) so lookups use
the same session, update the lookup and the later slash =
slash_by_source.get(...) usage to use that composite key, and when marking
matching tool_result entries for removal only include tool_result messages in
the same session whose visible text/content begins with "Launching skill:" (and
skip any tool_result that indicates an error/status != success if such meta
exists) instead of removing all tool_results with the same tool_use_id. This
ensures you only fold the intended slash body and drop the redundant "Launching
skill:" result within the same session.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 93790577-4afc-4767-906c-c6381acefe04

📥 Commits

Reviewing files that changed from the base of the PR and between 1dd392d and dc1770a.

📒 Files selected for processing (8)

• claude_code_log/factories/meta_factory.py
• claude_code_log/html/renderer.py
• claude_code_log/html/templates/components/message_styles.css
• claude_code_log/markdown/renderer.py
• claude_code_log/models.py
• claude_code_log/renderer.py
• test/__snapshots__/test_snapshot_html.ambr
• test/test_skill_pairing.py

[coderabbitai]

🧹 Nitpick comments (2)

claude_code_log/renderer.py (1)

2148-2175: Optional: pre-index tool_results for the inner scan.

PR description already calls out the O(K·N) inner search as a later perf pass. If you revisit it, a (render_session_id, tool_use_id) -> list[TemplateMessage] map built in the same sweep as slash_by_source turns the nested loop into a direct lookup, with the prefix / is_error guards unchanged. Skipping for now given the PR is already large and the Skill tool fan-out is small in practice.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@claude_code_log/renderer.py` around lines 2148 - 2175, The inner O(K·N) scan
over ctx.messages to find matching ToolResultMessage should be replaced by a
pre-indexed map so we can do a direct lookup: during the same sweep that builds
slash_by_source, also build a dict mapping (render_session_id, tool_use_id) ->
list[ToolResultMessage] (only include messages with message_index not None and
not is_error), then in the loop for each ToolUseMessage (the block that
references slash_by_source, consumed_indices, UserSlashCommandMessage) replace
the nested for other in ctx.messages scan with a lookup into that dict and
iterate that small list, applying the existing
_is_launching_skill_payload(other.content.output) check and adding
other.message_index to consumed_indices when matched.

claude_code_log/html/renderer.py (1)

427-436: Optional: mirror the Markdown renderer's empty-rendered guard.

MarkdownRenderer.format_ToolUseMessage returns content.skill_body verbatim when super() yields "" (see snippet from claude_code_log/markdown/renderer.py:676-690). Here the concatenation is unconditional, so a Skill ToolUseMessage whose params table renders to "" would emit a leading newline before body_html. Harmless in HTML, but the two renderers drift in behavior for the same content.

♻️ Optional parity tweak

         rendered = super().format_ToolUseMessage(content, message)
         if content.skill_body:
             body_html = render_markdown_collapsible(
                 content.skill_body,
                 "skill-body",
                 line_threshold=30,
                 preview_line_count=10,
             )
-            rendered = f"{rendered}\n{body_html}"
+            rendered = f"{rendered}\n{body_html}" if rendered else body_html
         return rendered

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@claude_code_log/html/renderer.py` around lines 427 - 436, The HTML renderer
unconditionally appends the collapsible skill body which causes a leading
newline when super().format_ToolUseMessage returns an empty string; update
claude_code_log/html/renderer.py in format_ToolUseMessage to mirror the Markdown
renderer's empty-rendered guard (check the local rendered variable returned by
super()) and only prepend a newline and append body_html when rendered is
non-empty, otherwise set rendered to body_html directly; reference rendered,
content.skill_body, format_ToolUseMessage, and render_markdown_collapsible to
locate the change.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@claude_code_log/html/renderer.py`:
- Around line 427-436: The HTML renderer unconditionally appends the collapsible
skill body which causes a leading newline when super().format_ToolUseMessage
returns an empty string; update claude_code_log/html/renderer.py in
format_ToolUseMessage to mirror the Markdown renderer's empty-rendered guard
(check the local rendered variable returned by super()) and only prepend a
newline and append body_html when rendered is non-empty, otherwise set rendered
to body_html directly; reference rendered, content.skill_body,
format_ToolUseMessage, and render_markdown_collapsible to locate the change.

In `@claude_code_log/renderer.py`:
- Around line 2148-2175: The inner O(K·N) scan over ctx.messages to find
matching ToolResultMessage should be replaced by a pre-indexed map so we can do
a direct lookup: during the same sweep that builds slash_by_source, also build a
dict mapping (render_session_id, tool_use_id) -> list[ToolResultMessage] (only
include messages with message_index not None and not is_error), then in the loop
for each ToolUseMessage (the block that references slash_by_source,
consumed_indices, UserSlashCommandMessage) replace the nested for other in
ctx.messages scan with a lookup into that dict and iterate that small list,
applying the existing _is_launching_skill_payload(other.content.output) check
and adding other.message_index to consumed_indices when matched.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 4405d23b-7975-46b3-b003-3f5922d44be8

📥 Commits

Reviewing files that changed from the base of the PR and between a584d7a and afb8e83.

📒 Files selected for processing (6)

• claude_code_log/factories/meta_factory.py
• claude_code_log/html/renderer.py
• claude_code_log/markdown/renderer.py
• claude_code_log/models.py
• claude_code_log/renderer.py
• test/__snapshots__/test_snapshot_html.ambr

✅ Files skipped from review due to trivial changes (2)

• claude_code_log/factories/meta_factory.py
• test/snapshots/test_snapshot_html.ambr

🚧 Files skipped from review as they are similar to previous changes (1)

• claude_code_log/markdown/renderer.py