refactor(tui): simplify pipeline event rendering logic#6
Closed
unohee wants to merge 9 commits into
Closed
Conversation
에이전트 실행 추적을 위한 핵심 모듈 구현: - Span/Trace 데이터 모델 정의 (trace_id, span_id, parent_span_id, status 등) - TraceCollector 클래스: start/end trace, start/end span, recordError - parent-child span 계층 구조 관리 - 22개 단위 테스트 전체 통과 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Replace claude-sonnet-4-20250514 (invalid) with claude-sonnet-4-5-20250929 across all configs - Add detailed stderr logging in planner.ts for debugging Claude CLI failures - Improve error handling in chat.ts to capture stderr output - Add real-time stderr logging and structured error messages This fixes 'Claude CLI exited with code 1' errors caused by invalid model ID. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Add /api/pipeline endpoint to expose stage events from eventHub - Redesign TUI Tasks tab to show pipeline stages in table format - Display task ID, stage name, model, tokens, cost, and status - Show 15 most recent pipeline events with status indicators (◐/●/✗) Tasks tab now matches web dashboard's pipeline view for better visibility. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Claude CLI does not support --disable-hooks flag. Removed to prevent 'unknown option' error. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Add 1-minute TTL cache for inProgress, backlog, and myIssues queries - Add clearLinearCache() to invalidate cache on mutations - Clear cache after updateIssueState and createIssue - Add cache logging for monitoring hit/miss rates This addresses Linear rate limit (5000 req/hour) by caching frequently accessed issue lists. Each heartbeat previously made 40+ API calls per 10 issues (N+1 query problem). Now only 1 call if cache is valid. Note: src/linear/linear.ts exceeds 1000 lines (1081), refactoring needed. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Implement token bucket rate limiter to prevent API abuse: - Claude API: 20 req/min (conservative, adjustable by tier) - Linear API: 80 req/min (4800/hour with 20% safety margin) - GitHub API: 80 req/min (4800/hour with 20% safety margin) Features: - Token bucket algorithm with automatic refill - Request queuing with configurable timeout (default 1-2min) - Per-service metrics tracking (total, rejected, queued, current tokens) - Graceful initialization/cleanup in service lifecycle - HTTP API endpoint: GET /api/rate-limits for monitoring Integration: - Applied to all Linear API calls (issues, states, mutations) - Ready for Claude CLI and GitHub API integration Note: src/linear/linear.ts needs refactoring (1084 lines) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Problem: - LanceDB append-only → 265 files accumulate (no deletion support) - 5.7MB with 8.4KB avg file size → fragmentation - cleanupExpired() only logs, doesn't delete - Backup files (.corrupted, .bak) accumulate (+1.6MB waste) Solution: - Implement full compaction: read → filter → deduplicate → recreate - Remove expired records (expiresAt < now) - Remove decayed records (decay >= 0.7) - Remove unimportant records (importance < 0.1) - Deduplicate similar memories (cosine similarity >= 0.85) - Clean up backup files Scheduling: - Daily at 2 AM (cron: 0 2 * * *) - Auto-check if needed (>20% waste or >1000 records) - Result: 265 files → ~50-100 files expected New functions: - compactMemoryTable(): Main compaction logic - shouldCompact(): Heuristic check - cleanupBackupFiles(): Remove .corrupted/.bak files Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Remove unused timestamp field from stage events - Simplify variable names for better readability (event -> ev) - Replace conditional chains with status map object - Merge task info extraction into single loop - Reduce code complexity while maintaining functionality Refs: INT-1088 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
unohee
added a commit
that referenced
this pull request
Apr 20, 2026
- longRunningMonitor: compile user-supplied completion patterns via `safeCompileRegex()` (length cap + try/catch) instead of `new RegExp()` directly, and reject malformed `checkCommand` on registration. The bash invocation is still intentional — dashboard operators need to compose shell probes — so it is annotated accordingly rather than rewritten. - scheduler: invoke `claude` via spawn argv instead of `bash -c "cd $path && claude -p \"$(cat $file)\" ..."`. The child process already gets `cwd`, so the extra shell layer (and the metacharacter-sensitive string interpolation) is unnecessary. - pairPipeline/longRunningMonitor: move tainted values out of the `console.*` first argument so they can never be interpreted as `util.format` specifiers. Resolves CodeQL `js/regex-injection` (#6, #7), `js/command-line-injection` (#9), and two `js/tainted-format-string` alerts (#22, #23).
unohee
added a commit
that referenced
this pull request
Apr 20, 2026
* ci: set minimal workflow permissions on CI pipeline
Adds a top-level `permissions: contents: read` to .github/workflows/ci.yml
so every job runs with the least-privileged GITHUB_TOKEN. Resolves
CodeQL `actions/missing-workflow-permissions` alerts 1-5.
* fix(web): strict CORS origin check and sanitized error responses
- Replace `origin.startsWith('http://tauri.localhost')`-style checks with
URL-parsed hostname matching. Prefix comparison admitted hosts like
`http://tauri.localhost.evil.com`; the new guard compares hostnames
exactly and validates the Tailscale CGNAT range by octet.
- Stop sending raw `String(err)` in JSON responses. `safeErrorMessage()`
returns only `err.message` (or a generic string), so stack traces and
non-Error object details no longer leak to the dashboard client.
Resolves CodeQL `js/incomplete-url-substring-sanitization` (#20, #21)
and `js/stack-trace-exposure` (#10–#19).
* fix(automation): harden regex, shell, and logging inputs
- longRunningMonitor: compile user-supplied completion patterns via
`safeCompileRegex()` (length cap + try/catch) instead of `new RegExp()`
directly, and reject malformed `checkCommand` on registration. The
bash invocation is still intentional — dashboard operators need to
compose shell probes — so it is annotated accordingly rather than
rewritten.
- scheduler: invoke `claude` via spawn argv instead of
`bash -c "cd $path && claude -p \"$(cat $file)\" ..."`. The child
process already gets `cwd`, so the extra shell layer (and the
metacharacter-sensitive string interpolation) is unnecessary.
- pairPipeline/longRunningMonitor: move tainted values out of the
`console.*` first argument so they can never be interpreted as
`util.format` specifiers.
Resolves CodeQL `js/regex-injection` (#6, #7), `js/command-line-injection`
(#9), and two `js/tainted-format-string` alerts (#22, #23).
* fix(knowledge): defend graph file paths and untaint log calls
- store.ts: resolve every `<STORE_DIR>/<slug>.json` path through
`resolveGraphPath()` and reject anything that escapes STORE_DIR.
`toProjectSlug()` already strips non-alnum chars, so this is
defense-in-depth for callers that forget to sanitize.
- store.ts / index.ts: pass slugs and project paths as separate
`console.*` arguments instead of interpolating them into the first
argument, so `util.format` never sees tainted specifiers.
Resolves CodeQL `js/path-injection` (#28) and the remaining
`js/tainted-format-string` alerts (#24–#27).
* fix(monitor): argv-only checkCommand and regex sanitizer
BREAKING CHANGE: LongRunningMonitor.checkCommand is now `string[]`
(argv) instead of a shell string. Pipes, redirects, and substitutions
no longer work out of the box — wrap shell logic in a script and call
the script via argv.
- Replace `execFile('bash', ['-c', command])` with
`execFile(program, args, { shell: false })`. Shell metacharacters in
argv elements are inert because no shell interprets them, which
eliminates the underlying command-injection surface and resolves
CodeQL alert `js/command-line-injection`.
- Validate every argv token against a control-character-free allowlist
(`ARGV_SAFE`) and bound length at register time and inside
`executeCheck`. Legacy persisted monitors with string checkCommand
are skipped at load with a warning rather than crashing the service.
- Zod schema (`LongRunningMonitorConfigSchema.checkCommand`) and the
`POST /api/monitors` handler now require a non-empty string array,
with a clearer 400 error pointing at the new shape.
- Regex sanitizer: `safeCompileRegex()` additionally rejects any input
containing control characters via an ALLOWED_REGEX_CHARS allowlist.
This doubles as a CodeQL sanitizer for `js/regex-injection`.
- Refresh `config.example.yaml` monitor example to argv form with a
note about shell semantics.
Resolves the two new alerts (#29 `js/regex-injection`,
#30 `js/command-line-injection`) flagged on PR #46.
* fix(monitor): constrain checkCommand program to an allowlist
Argv-only invocation alone still let `execFile` spawn any binary
reachable from `PATH`, which CodeQL correctly flags as a
command-injection sink. Gate the program before it reaches `execFile`:
- Bare names must be in ALLOWED_PROGRAMS (curl, wget, ssh, jq, grep,
awk, sed, cat, tail, head, nvidia-smi, kubectl, docker, podman).
Extending the list is a deliberate code change, not a config change.
- Absolute paths / `~/...` must resolve under a trusted prefix
(`/usr/local/bin/`, `/usr/local/sbin/`, `/opt/`, `$HOME/bin/`,
`$HOME/.local/bin/`, `$HOME/scripts/`, `$HOME/.openswarm/monitors/`).
- `..` is rejected outright to prevent traversal tricks against the
prefix checks.
- Validation runs both inside `isValidArgv` and again at the call site
in `executeCheck`, so the program value flowing into `execFile` is
always vetted — this is what CodeQL needs to see as a sanitizer.
Resolves the remaining CodeQL alert #31 `js/command-line-injection`
on PR #46.
Link-Start
pushed a commit
to Link-Start/OpenSwarm-unohee
that referenced
this pull request
Jul 3, 2026
Behavioral + deterministic guardrails against the systematic swarm defects surfaced reviewing autonomous PRs unohee#201-207 (approve 3 / request-changes 3 / reject 1). Prompt rules (worker Rules + reviewer criteria, en + ko): - unohee#1 on import/SDK failure, fix the env — never re-author a third-party package or spoof its __version__ - unohee#2 cite the counterparty's real producer/consumer code for cross-service contracts; no self-referential tests - unohee#4 don't delete "verified" statements without counter-evidence; include a before/after distribution for score/gate/metric changes - reviewer criteria: contract integrity, evidence & wiring, scope Deterministic guards (pipelineGuards, non-blocking): - deadModuleCheck (unohee#5): flag newly-added source modules nothing imports/calls - reformatCheck (unohee#6): flag reformat-only files and oversized diffs - getWorkingDiffDetail (gitTracker): per-file numstat + new/whitespace-only flags - surface non-blocking guard warnings to the reviewer (previously log-only) Wired via PipelineGuardsConfig (deadModuleCheck/reformatCheck) + config.example. Out of scope, tracked as follow-up: canonical-package registry (unohee#1), deterministic contract verification (unohee#2), orchestrator file-overlap serialization + PR-overlap report (#3).
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
1 participant
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
Refactor pipeline event rendering logic in Chat TUI for better code quality and maintainability.
Changes
Files Changed
src/support/chatTui.ts- Simplified pipeline event rendering (39 insertions, 76 deletions)Related Issues
🤖 Generated with Claude Code