nhangen-tools

Obsidian vault integration for Claude Code. Auto-capture git commits, save sessions, query past work with /recall, export conversations. Supports user-defined VAULT.md conventions.

11 skills · 8 commands · 2 agents · hooks

Autonomous CEO agent. Reads Obsidian vault, dispatches 6 specialized subagents, 8 playbooks, 7 skills, cron execution pipeline with three-phase tier enforcement. CLI with setup/doctor/test.

8 skills

Dispatch long-running research tasks to the Manus AI agent platform from Claude Code via slash command. SessionStart + Stop hooks surface newly-completed tasks both on next session and mid-conversation (zero extra Claude turns). Retrieves file deliverables from their signed CDN links, so a task whose payload is an attachment no longer looks complete with the payload missing. On-demand polling, optional Obsidian filing, explicit-invocation cost gating.

3 commands · hooks

Source-grounded project glossary. Auto-invokes when the agent encounters an unfamiliar acronym or project-specific term in a registered repo: resolves the current cwd to a domain via per-machine config (with git-worktree fallback), reads the matching glossary, and answers cited rather than guessing. Also handles seed (build candidates from commits / GitNexus / docs) and drift-check (verify every citation still resolves).

1 skill · 1 command

Behavioral markdown documentation scanner. Analyzes session history to find gaps in CLAUDE.md, rules, memory, and project settings. Detects repeated file reads, command failures, recurring user corrections, allowlist gaps, unused CLAUDE.md sections, documented-but-still-re-read paths, and Cursor↔Claude rule drift. Guided walkthrough to approve/skip/defer recommendations.

1 skill · hooks

Auto-compact when context fill crosses thresholds. Stop-hook gate computes fill % from the live transcript; fires an advisory at 35% and escalated nudge at 50%, both directing /compact. Context-window divisor is a dynamic var (CONTEXT_LOOP_WINDOW override; falls back to Opus 1M / other 200K heuristic).

1 skill

Token spend analytics and cost alerts for Claude Code. 16 report modes including weekly credit consumption against your plan cap, cache intelligence, tooling analysis by layer, per-artifact Write/Edit cost analytics, and a credit-denominated auto-checkpoint hook.

1 skill

PreToolUse hook that runs gitnexus impact on the symbol being edited and injects the blast-radius summary as additionalContext at the moment of editing. Companion to GitNexus's shipped Grep/Glob/Bash hook (which only covers the search phase, not the edit phase).

Captures (features, label, evidence) tuples from Claude Code sessions as ML training data. PreToolUse/PostToolUse hooks write per-tool-call feature rows; phrase-based detector + LLM-as-judge label corrections against a failure-mode taxonomy seeded from ~/.claude/rules/. Joiner emits JSONL ready for fine-tuning, DPO, or eval. Includes `pt cost-by-failure` fusion view joining findings to token-scope per-session cost. Dependency-free (stdlib-only, no venv). Private repo.