Ecosystem Update — 2026-04-02

April 2, 2026 · generated by the ecosystem-update Claude Skill

TL;DR

All isolation-eligible agents now have isolation: worktree — isolation coverage complete
type: prompt hooks confirmed by community — allows inline LLM-as-validator hooks with no Python script required; verify syntax before wiring
Build Queue additions: Dippy (AST bash auto-approve), claude-rules-doctor (dead-rule detector), claude-tmux

Quick Wins

(None this run — agent isolation complete, remaining items require new scripts/installs)

Build Queue

Dippy (hook) — github.com/ldayton/Dippy — AST-based bash auto-approve that semantically classifies safe vs destructive commands; complements pre_tool_guard.py's regex approach with actual syntax-tree analysis
claude-rules-doctor (tool) — github.com/nulone/claude-rules-doctor — Detects dead .claude/rules/ globs by validating patterns match files; wire to SessionStart or run periodically as CLAUDE.md grows
claude-tmux (mcp/client) — awesome-claude-code catalog — Manage Claude Code sessions within tmux; pairs with existing worktree + parallel-agent patterns
PermissionDenied hook (hook) — coleam00/claude-code-new-features-early-2026 — New hook event (not in current settings.json); fires when permission is denied; verify event name stability before wiring script
TDD Guard (hook) — github.com/nizos/tdd-guard — PostToolUse hook blocking non-TDD file changes (carried from previous run)
cc-devops-skills (skill) — github.com/akin-ozer/cc-devops-skills — IaC-focused DevOps skills with validators, generators, CLI tools (carried from previous run)

Research

type: prompt hook type — gist community — Hooks can use "type": "prompt" to run inline LLM-as-validator without a Python script; Impact=2, Effort=1 — verify exact settings.json syntax against official docs before implementing. If confirmed, can replace some lightweight Python guard scripts.
type: prompt hooks on Stop / PostToolUse — Could add natural-language verification at Stop ("did this complete all acceptance criteria?") with zero new scripts; pending syntax verification
Session Teleportation / --teleport — built-in CLI flag (carried)
Dynamic !command injection in SKILL.md — embed shell output at skill load time (carried)

Already Have

isolation:worktree (worker, planner, reviewer, python-reviewer, typescript-reviewer), sandbox:read-only (validator), PostCompact hook, PreCompact hook, PermissionRequest hook, PostToolUseFailure hook, TaskCompleted hook, SubagentStop hook, SessionStart hook, UserPromptSubmit hook, Stop hook, pre_tool_guard.py, classify_prompt.py, permission_audit.py, agent teams (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1), claude-mem MCP, playwright MCP, plan mode (planning-gate skill), parry, Trail of Bits security skills, compound-engineering-plugin, context-engineering-kit, /batch built-in, /loop, /schedule

Rejected

parry — removed 2026-04-02: false-positive taint blocked all tools session-wide; more harmful than protective at current sensitivity level
container-use (dagger) — alignment failure: git worktrees already provide agent isolation; Docker-based isolation requires a new service dependency without proof worktrees are insufficient
TypeScript Quality Hooks — covered by typescript-reviewer agent
cchooks Python SDK / claude-hooks TypeScript SDK — existing Python scripts work without an SDK layer
Claudio sounds hook — minor convenience only
Britfix — irrelevant to current workflow

Sources checked: github.com/hesreallyhim/awesome-claude-code, howborisusesclaudecode.com, github.com/shanraisshan/claude-code-best-practice, github.com/coleam00/claude-code-new-features-early-2026, community gists Tier 2 fetched: no (within 24h window) Tier 3 fetched: no (within 7d window) Run at: 2026-04-02T15:45:00Z