Timeline UX + multi-file memory viewer + captain-hook v2.1.92 expansion#55
Merged
JayantDevkar merged 15 commits intomainfrom Apr 12, 2026
Merged
Timeline UX + multi-file memory viewer + captain-hook v2.1.92 expansion#55JayantDevkar merged 15 commits intomainfrom
JayantDevkar merged 15 commits intomainfrom
Conversation
…g [skill_name] segment
Collaborator
Author
|
Five HIGH-severity issues plus supporting cleanups and tests. HIGH - ToolCallDetail: remove 'content' from skipKeys — the Write tool content preview was filtered out before the render path could read it, so feature #2 of PR #55 silently did not work. Kept 'task_subject' in skipKeys since it's read directly from event.metadata. - SessionCard + projects/[project_slug]: extract getSessionUrlIdentifier helper so the last-opened-highlight comparison uses the same slug/uuid resolution as SessionCard's href, fixing highlight loss on sessions where liveSession.slug differs from session.slug. - markdownCopyButtons: query h1/h2/h3 (was h2/h3); h1 and h2 always get a button, h3 keeps the 150-char gate. Matches the PR description. Removed stale h2-vs-h3 visual-distinction comment. - ToolCallDetail TaskUpdate: restore input.subject to hasChanges and add a "Rename to" row inside the Changes panel so a subject-only rename is visible as an explicit delta (header still shows inherited subject). - conversation_endpoints: tighten regex from r"Task #(\w+)" to r"Task #(\d+)" and move `import re` to module top. Added 3 regression tests in a new test_conversation_endpoints.py covering happy path, no matching TaskCreate, and unparseable result content. MEDIUM - markdownCopyButtons: delete dead `cleanupFns` array — it was populated but never invoked; DOM removal handles listener cleanup via GC. - jsonl_utils: replace copy.deepcopy(base) with {**base} shallow copy. The merge function only mutates the `message` key which is already rebuilt as a fresh dict, so deepcopy was wasted allocation for base64-image payloads. Drop `import copy`. - test_jsonl_utils: add TestMessageMerging with 8 tests — 4 direct unit tests of _merge_user_message_dicts (image-source drop, real-extra preservation, empty extra, legacy content key) and 4 integration tests via iter_messages_from_jsonl (same-timestamp merge, different-timestamp no-merge, cross-type no-merge, 3-way merge). - sessions/+page.svelte: make restoreScroll async and await tick() before requestAnimationFrame so scroll restore lands on fully-rendered DOM. LOW - ToolCallDetail: strip trailing newline before counting lines in the Write content preview so "a\nb\n" reads as 2 lines not 3. Verification - api: pytest tests/test_jsonl_utils.py tests/test_conversation_endpoints.py tests/test_session.py tests/test_agent.py — 149 passed. - frontend: svelte-check clean on all touched files (pre-existing errors only in skills/[skill_name=skill_name]/* from the PR's own route rename). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Owner
|
@the-non-expert — pushed code-review fixes in 🔴 HIGH (5 fixes)
🟡 MEDIUM (4 fixes)
🟢 LOW
Verification
Manual test plan — please verify before merging
Full diff: https://github.com/JayantDevkar/claude-code-karma/pull/55/files |
Post-audit regression fix for the merge logic added in the previous
commit. Claude Code v2.1.83 changed the image-attachment marker from
`[Image: source: /var/folders/...]` to `[Image #N]`, with a trailing
space added in v2.1.85+. The existing `startswith("[Image: source:")`
check missed the new format, so `_merge_user_message_dicts` would leave
the redundant marker text in the merged content for any session created
by a Claude Code version released after 2026-03-25.
Changes
- Extract a `_is_image_marker_text` helper that matches both prefixes.
- Update the docstring to call out the v2.1.83 + v2.1.85 format variants
so the next schema change is easier to spot.
- Add `test_merge_drops_image_hash_number_marker` covering a 2-image
attachment with both `[Image #1]` and `[Image #2] ` (trailing space)
marker variants in the extra message.
Verification: pytest tests/test_jsonl_utils.py → 18 passed.
Cite: Claude Code changelog v2.1.83 (2026-03-25) and v2.1.85 (2026-03-26).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
….1.92
Expand the captain-hook Pydantic library with 11 new hook event types
introduced between Claude Code v2.1.83 and v2.1.92, bringing the total
from 13 to 24 supported hooks.
New hook classes:
Context:
- InstructionsLoadedHook (CLAUDE.md / rules file loaded)
User interaction:
- PermissionDeniedHook (auto mode denied a tool call; can request retry)
- ElicitationHook (MCP server requests structured input)
- ElicitationResultHook (user response to MCP elicitation)
Filesystem (new module fs_hooks.py):
- CwdChangedHook (working directory changed)
- FileChangedHook (external file change detected)
Agent Teams (new module team_hooks.py, experimental):
- TaskCreatedHook
- TaskCompletedHook
- TeammateIdleHook
Worktree (new module worktree_hooks.py):
- WorktreeCreateHook (HTTP hook can override worktreePath)
- WorktreeRemoveHook
Output model changes:
- PreToolUseOutput.permission_decision now accepts "allow", "deny",
"ask", and "defer" (the new "defer" value supports the headless
`-p --resume` pause/resume flow).
- New PermissionDeniedOutput with a single retry: bool field for
requesting that Claude retry a denied tool call.
Other changes:
- HookEventName Literal in base.py extended with all 11 new event names.
- HOOK_TYPE_MAP, HookEvent union, and __all__ in src/captain_hook/__init__.py
extended with the new classes.
- Backward-compat shim models.py mirrors the new exports.
- 107 new tests added (fixtures, parser dispatch, round-trips, output
models): 131 -> 238 total tests, all passing.
- README.md and CLAUDE.md updated to reflect the 24-hook count and
the new module layout.
- 11 new docs/hooks/*-info-available.md files following the existing
template.
No existing hook classes or test cases were modified.
14 action items across 3 sprints tracking dashboard updates needed for Claude Code releases between 2026-03-20 and 2026-04-06. - Tier 1 (Blockers): JSONL merge regression (shipped), captain-hook expansion (11 new types), Teams discovery, bare mode detection - Tier 2 (High Priority): PowerShell/Teams tool recognition, frontmatter parsing (initialPrompt, shell, paths), managed-settings fragment merging, MCP metadata preservation, 14 new settings fields - Tier 3 (Experimental): Extended thinking visibility, hook conditional if support, session resumption hardening Cite: claude-code-guide audit (2026-04-06)
Owner
|
@the-non-expert — two more commits pushed on top of the earlier review fixes. The scope of this PR grew beyond timeline UX because I ran a Claude Code release audit (v2.1.81→v2.1.92) in the same session and chose to keep all follow-up work together per @JayantDevkar's preference rather than splitting into 3 PRs. Current commit stack on
|
| Suite | Count | Result |
|---|---|---|
pytest api/ |
1444 passed, 6 skipped | ✅ |
pytest captain-hook/ |
238 passed (baseline was 131) | ✅ |
| Svelte-check on touched files | clean | ✅ |
What I'd like you to focus on in review
- Timeline UX changes (your original work) — already reviewed by me in the earlier comment
- H13/H4/H10/H6/H1 review-fix commit (
13cd5bf) — needs manual test on the Write content panel and the last-opened session highlight [Image #N]regression (272f506) — purely additive pattern match, backed by a new test- captain-hook expansion (
cf7eb26) — please sanity-check the new hook models against the official Claude Code hook docs. The schemas are my best-effort interpretation of the audit report, and any field that was ambiguous I defaulted toOptional[str]to be permissive. The 3 sub-sections marked{needs clarification}in the audit plan correspond to exactly these ambiguities. - audit plan doc (
e122b3c) — content review only, it's a planning doc not code
Rollback strategy if anything is wrong
Each commit is independently revertable. The 4 new commits on top of your original 4 are ordered from lowest-risk (review fixes) to highest-risk-of-subjective-disagreement (audit plan). Revert from the top down if needed.
Scratch branches cleaned up
The work landed here via cherry-pick from two scratch branches (feat/captain-hook-v2.1.92-events and worktree-agent-a380b18f). Both are being deleted locally now that they're merged here. No remote scratch branches were pushed.
Design for reworking MemoryViewer to handle the new auto-memory layout (MEMORY.md index + topical child files with YAML frontmatter). Reader- first UX: hover previews (Wikipedia-style) for link metadata, side panel drawer for full-content reading, collapsible orphan section for files not referenced by the index. Covers API changes (2 endpoints), component split, backwards compatibility, and test strategy. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace the single-file /projects/{name}/memory response with a
{index, files} shape that enumerates every *.md in the project's
memory/ directory. Add /projects/{name}/memory/files/{filename}
for fetching individual child file content on demand.
- New schemas: MemoryIndexEntry, MemoryFileMeta, ProjectMemoryFileResponse
- Hand-rolled YAML frontmatter parser (no new dependency) with graceful
fallback on malformed blocks
- Markdown link extraction computes linked_from_index for each child
- Strict path validation on the per-file endpoint: regex allowlist,
reject path separators / .. / leading dot / null byte, plus a
resolve() + is_relative_to() containment check for symlink escapes
- Backwards compatible: projects with only MEMORY.md return files=[]
- 30 new tests covering all spec error cases and 12 path-traversal variants
Design: docs/features/2026-04-07-multi-file-memory-ui.md
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Rework the project memory UI to handle the new auto-memory layout
(MEMORY.md index plus topical child files with YAML frontmatter).
Reader-first UX: the index narrative is the page. Links within the
index are rewritten to in-app interactive elements — hover shows a
Wikipedia-style popover with the target file's metadata (type, word
count, modified time), click opens the file in a right-side drawer.
Orphan files (present on disk but not referenced from MEMORY.md)
surface in a collapsible section at the bottom.
- MemoryViewer.svelte: refactored into a shell that owns state,
fetches the new {index, files} response, and debounces hover 150ms
- MemoryIndex.svelte: renders MEMORY.md through the existing marked +
DOMPurify pipeline, applies the rewriteMemoryLinks action
- MemoryHoverCard.svelte: floating popover with manual viewport flip
- MemoryFilePanel.svelte: bits-ui Dialog styled as a right-side sheet,
fetches /memory/files/{filename} on open, handles 400/403/404/network
errors with a Retry button, swaps content when filename changes
- MemoryOrphanList.svelte: collapsible "Other memory files" section
- rewriteMemoryLinks.ts: Svelte 5 action that post-processes the
rendered DOM, matches a[href$=".md"] anchors against known files,
attaches hover/click listeners, and marks broken links visually
- api-types.ts: replaced ProjectMemory with the new {index, files}
shape, added MemoryFileMeta, ProjectMemoryFile, MemoryFileType
- vite.config.ts: added svelteTesting() plugin (VITEST-gated, no-op
in production) to enable Svelte 5 component tests
- 37 new tests across 4 files covering loading/error/orphan states,
DOM rewriting lifecycle, hover card rendering, and panel refetch
Integration point at +page.svelte:1725 unchanged.
Design: docs/features/2026-04-07-multi-file-memory-ui.md
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
MemoryFilePanel.fetchFile() mutated shared state (fileData, bodyFading, loading, error) without tracking which invocation was current. This created three reachable bugs with a single root cause — no fetch lifecycle management: 1. Out-of-order resolution: click A → click B → A's fetch resolves after B's → fileData ends up as A (wrong file shown permanently). 2. bodyFading race: A's 80ms setTimeout could fire while B was still loading, briefly showing A without the fade-out covering it. 3. Close-during-fetch leak: closing the panel mid-fetch cleared state, but the in-flight fetch still wrote fileData on resolve, leaking stale content into the next open. Fix: add a generation counter (fetchGen). Each fetchFile captures ++fetchGen at entry and checks `myGen !== fetchGen` after every await and in the finally block. Superseded invocations drop their state writes silently. Panel close also bumps fetchGen to invalidate any in-flight fetch. Chose a gen counter over AbortController for simplicity — the bandwidth saved by actual cancellation is negligible for a localhost API, and the code stays in one mental model (no try/catch on AbortError, no extra imports). Adds a regression test that resolves an older fetch AFTER a newer one and asserts the newer file's content remains visible. Test fails against the pre-fix code at the exact expected assertion. Verification: 261 frontend tests pass (up from 260), svelte-check clean, production build succeeds. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
… components Addresses two HIGH quality findings from the code review of the multi-file memory UI feature. No behavioral change — verified by the existing 261-test suite including the race regression test. 1. Reuse existing renderMarkdownEffect helper MemoryIndex.svelte and MemoryFilePanel.svelte both reimplemented the marked + DOMPurify pipeline inline (the same ~12-line block each), when utils.ts already exports renderMarkdownEffect — a helper used by 4 other routes (agents, about, commands, skills). Two copies of a security-critical sanitization pipeline drift over time; centralize on the existing helper. 2. Extract shared type badge module TYPE_BADGE_CLASSES, TYPE_LABELS, badgeClass, and badgeLabel were duplicated byte-for-byte across MemoryHoverCard, MemoryFilePanel, and MemoryOrphanList (~25 lines each × 3 = ~75 lines). Extract to a single memoryTypeBadge.ts module. MemoryOrphanList's badge needed `shrink-0` because it lives inside a flex row — accept optional extra classes via a second parameter to badgeClass() rather than forcing per-call string concatenation at every site. Net: -84 lines (+18 / -102) across 4 edited files + 1 new module. Verification: 261 tests pass, svelte-check clean (0 errors), build succeeds. The fetch-race regression test from the previous commit continues to pass, confirming the refactor did not reintroduce any of the bugs. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Fixes ruff F401 CI failure on PR #55. `pytest` was imported but never referenced — the tests use plain assert statements without fixtures. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
JayantDevkar
changed the title
…utils tests Fixes Python Lint CI failure on PR #55. `ruff format --check` flagged two test files with whitespace drift; applying the formatter is idempotent and matches the repo's existing ruff 0.15.9 config. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
JayantDevkar
approved these changes
Apr 12, 2026
Owner
JayantDevkar
left a comment
JayantDevkar
left a comment
There was a problem hiding this comment.
Self-approving as repo owner. Local verification: 1474 api + 238 captain-hook + 261 vitest + svelte-check + ruff all green. CI green on dbf5d8f after the formatter fix.
JayantDevkar
approved these changes
Apr 12, 2026
Owner
JayantDevkar
left a comment
JayantDevkar
left a comment
There was a problem hiding this comment.
Self-approving as repo owner.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR started as timeline UX polish and grew into four themed chunks, kept bundled per the 2026-04-07 scope decision. All 13 commits pass CI locally (ruff, pytest, svelte-check, eslint, vitest).
What's in the PR
1. Timeline UX (original scope) — 4 commits
markdownCopyButtons.ts).TaskCreatetool results into ataskId → subjectmap; TaskUpdate events display the task description instead of justTask #N./sessionsand/projects/[slug](sessionStorage + beforeNavigate +await tick()+ rAF).[skill_name]segment.2. Code-review fix commit (
13cd5bf)Addresses 5 HIGH + 4 MEDIUM + 1 LOW findings from PR review — summary inline in the commit and the earlier PR comment. Key fixes: Write-tool content panel un-skipped, last-opened highlight drift between live/canonical slugs, H1 selector + code-block gating, TaskUpdate rename delta,
Task #(\d+)regex tightening,deepcopydropped from jsonl_utils merge.3. JSONL
[Image #N]marker fix (272f506)Claude Code v2.1.83+ serializes attached images as
[Image #N]placeholders. JSONL merge now recognizes and preserves the pattern. +1 regression test.4. Captain-hook v2.1.92 expansion (
cf7eb26)Eleven new hook models covering Claude Code v2.1.83–v2.1.92:
InstructionsLoaded,CwdChanged,FileChangedPermissionDenied,Elicitation,ElicitationResultTaskCreated,TaskCompleted,TeammateIdleWorktreeCreate,WorktreeRemovePlus
"defer"added toPreToolUseOutput.permissionDecision(v2.1.89) and a newPermissionDeniedOutputclass withretry: bool.HOOK_TYPE_MAPgrows from 13 → 24. Adds 3 domain modules (fs_hooks.py,team_hooks.py,worktree_hooks.py) and 11 reference docs. +107 tests (238 total, up from 131).5. Audit remediation plan (
e122b3c)docs/features/2026-04-06-claude-code-audit-action-items.md— 874-line planning doc tracking 14 action items across Blockers / High Priority / Experimental tiers for the v2.1.81→v2.1.92 remediation sprint. Docs-only change.6. Multi-file project memory (5 commits)
1b4634f): new endpoints onrouters/projects.pyfor list/read/write/delete of.claude/CLAUDE.mdsiblings. +448-line test suite intest_memory.py.4f13259):MemoryViewersplits intoMemoryIndex,MemoryFilePanel,MemoryOrphanList,MemoryHoverCard, plus arewriteMemoryLinks.tsaction that intercepts cross-file links and routes them through the viewer without a page load. +8 vitest suites.d770c15): sequence-gate onMemoryFilePanelfetch resolution to prevent a late response from clobbering a newer selection. +1 test.d5a9652):memoryTypeBadge.tsextracted soMemoryIndex,MemoryFilePanel,MemoryHoverCard,MemoryOrphanListshare a single source for type badges and the markdown pipeline.a3e9b54):docs/features/2026-04-07-multi-file-memory-ui.mdarchitecture spec.7. Ruff lint fix (
d51e389)Removes unused
pytestimport fromtest_conversation_endpoints.pyso Python Lint CI goes green.Verification
api/ pytestcaptain-hook/ pytestapi/ ruff checkfrontend/ svelte-checkfrontend/ eslintfrontend/ vitestManual test plan
/projects/[slug], press back, confirm the card you came from glows with the ring#input.subjectis present/sessionslands on the right page and scroll positionRollback strategy
Each commit is independently revertable. Ordered from lowest-risk (lint fix,
[Image #N]pattern match) to highest-risk-of-subjective-disagreement (audit plan, multi-file memory UX). Revert from top down if needed.🤖 Generated with Claude Code