Plugins

Guarded BI readiness workflow and reviewed public knowledge for Claude Code.

from awudevelop/claude-plugins

Plugin รวม Skills สำหรับจัดการ Session ของ Claude Code — ตั้งชื่อ บันทึกวัตถุประสงค์ รักษาบริบทของ session และเขียนเอกสารส่งไม้ต่อตอนจบ session

from cadrianmae/claude-marketplace

Claude Codeの過去の会話ログを発掘・分析し、開発の試行錯誤プロセスやインサイトを構造化して要約する

Archive Claude Code session transcripts to local and remote targets (S3, rsync/NAS, or any command) on session end — per-machine configurable, nothing hardcoded

Inspect, resume, and hand off agent sessions across Codex CLI and Claude Code. `session-bridge` CLI reads both session stores; resume-session (read) + handoff-session (switch tools) skills.

Track Claude Code's 5-hour session usage from local JSONL and pause tool calls before hitting the limit.

Hooks log events, /remember searches them, /carto opens the Explorer, /focus orients on a project

Report token usage, wall-clock time, and per-model cost for Claude Code sessions from the on-disk ~/.claude/projects JSONL transcripts — grouped by session or git worktree, with a compare mode. Dedups streamed/re-serialized usage and prices per model so totals reconcile with ccusage. Read-only.

Report token usage, wall-clock time, and per-model cost for Claude Code sessions from the on-disk ~/.claude/projects JSONL transcripts — grouped by session or git worktree, with a compare mode. Dedups streamed/re-serialized usage and prices per model so totals reconcile with ccusage. Read-only.

다른 Claude 세션에 일을 맡기고 결과를 받는 절차 — 이미 떠 있는 세션(SendMessage 후 턴 종료, 폴링 금지), herdr 로 새로 띄우는 세션, 다른 머신의 Remote Control 세션이 각각 다른 경로. 주소는 ListAgents 의 이름이고 관측자마다 달라지므로 --remote-control <이름> 으로 못박는다. herdr 로 조작할 세션은 --settings '{"tui":"default"}' 로 띄운다 — 풀스크린은 alternate screen 이라 pane read 가 현재 화면 이상을 못 읽는다(입력은 되고 읽기만 막혀 증상이 헷갈린다). 세션 교체(RC 끊기 → /exit → pane 재사용), 새 머신 부트스트랩, 회신을 그대로 옮기지 않고 검증하는 절차 포함.

File-explorer-style session management for Claude Code

Cross-platform session historian agent. Searches prior Claude Code and Codex session transcripts for what was tried, what failed, and what was decided in earlier sessions on the same repo. Invoked by other skills (/compound, /xplan, /debug) to surface institutional knowledge a fresh session cannot see.

from x85446/claudecodetricks

Analyze Claude Code sessions to find token waste, linter loops, and missing automations, then recommend and apply fixes

Analyze conversation for friction signals (corrections, repeated instructions, deviations) and interactively update project documentation (.claude/rules/, CLAUDE.md) with verified learnings. Coordinates with Claude Code's auto memory to avoid duplication.

Structured session shutdown via /sds. Autonomous wrap-up sweep: commit dirty work, update referenced issues, run /reflect, write a handoff, broadcast to sibling clones. Complements /startup at the other end of the session.

Per-project Claude Code session log. Injects recent history at session start and nudges Claude to write entries after meaningful work.

from JGillam/practical-cc

Slash commands to hand off / pick up Claude Code sessions through Obsidian. /session-management:handoff writes everything worth preserving to the project's Obsidian folder as Diátaxis-structured pages. /session-management:pickup reads it back at session start.

Adds commands to manage a session in the project, /start-session to start a session and /summarize-session to summarize the session and update the summary.md file.

Per-turn token, cost, and cache analytics for Claude Code sessions. Multi-format export (text/JSON/CSV/MD/HTML) with session blocks, weekly roll-up, subagent attribution, dynamic-workflow tracking, and 9-category turn waste classification. HTML reports include cost timeline, hour-of-day punchcard, model-switch dividers, cache-break badges, a per-turn detail drawer, and a per-request breakdown that groups turns by the prompt that drove them. Sibling skills: audit-session-metrics audits exports for waste patterns (summarisation-only — run on /model haiku for ~10× cheaper); task-breakdown groups requests into semantic tasks with worth-it / wasted verdicts and a Tasks companion page. Zero network, stdlib-only Python.

Mine Claude Code sessions for knowledge extraction. Process session history to identify learnings, patterns, and improvement opportunities.

Records the complete session content and logs it to a daily work directory with a dynamic filename based on the active CLI agent. Use this for automated progress tracking and documentation.

Generate an explorable HTML report of Claude Code session usage — tokens, cache efficiency, subagents, skills, and the most expensive prompts — from local ~/.claude/projects transcripts.

Interactive session retrospectives — deterministic Stop/PreCompact triggers, diff-driven interview, native memory entries. No external services.

Start-of-day and end-of-day routines for Claude Code: /wind-down writes one consolidated brief across your sessions, /good-morning reads it back and re-checks what changed overnight.

Check session security implementation

Eight skills for working inside one conversation. braindump plays back a pasted brain dump in very simple words and waits for the owner's yes before any work starts. explain-simply says the last answer again as plain bullets without losing a number, date, path, or name. grill-me interviews the owner one question at a time and checkpoints every answer to a persistent file before continuing. handoff runs the short persistent-information review, then drafts a prompt a fresh session can start from, opening with the goal and checked by a second agent that never saw the conversation. session-summary tables what the owner asked for and gives each request an honest status. track-tasks keeps every still-open topic on Claude Code's built-in task list, including parked topics, unanswered questions, and blocked work. spec-check reads the specification an agent is about to build from or design a solution from, flags anything that could skew the work, and proposes fixes the owner approves before building starts. unslop takes writing that already exists, names every pattern in it that reads as machine-written, and hands back a version with those patterns gone and a voice put back, writing to a file only on the owner's yes.