Claude Code ships with Bash, Read, Write, Edit, Glob, Grep, WebFetch, Task, Plan. For a single agent on a single task, they're fine. Once you're running a multi-agent system, the defaults break fast.
The problems we hit:
1) No cross-repo exploration. Want an agent to read another project's code? You need to manually configure permissions. There's no "go explore this OSS repo and answer my question."
2) Summarized web fetching. WebFetch is actually a subagent that summarizes a single page into a haiku-length response. Can't trace links or browse related pages. Fetches fresh every time — no caching.
3) Text-level editing. Edit has fuzzy matching, which helps — but it's still raw text. When tree-sitter can give you an AST with named symbols, why make the model reproduce strings to target a function?
4) Ephemeral tasks and plans. Tasks don't persist outside the session. Plans vanish when context resets. Neither supports multi-round review or structured editing.
5) Flat delegation. The Agent tool can't nest — an agent spawned by Agent can't spawn its own sub-agents. One level of delegation, period. A review leader can't spawn specialized sub-reviewers.
6) No isolation. Bash runs on your host. No sandboxing, no filesystem allowlists. You either yolo it or do annoying permission work.
What we built instead:
- ttal ask: multi-mode research (--url caches pages locally with 1-day TTL and follows links, --repo auto-clones OSS repos, --project spawns read-only sandboxed agents in the right directory). Runs on logos, a pure-bash agent loop — works with any LLM provider, so you can use cheaper models for exploration.
- organon (src/url/search): tree-sitter AST editing. See a symbol tree, pick a 2-char ID, pipe new code via heredoc. No text matching. Same pattern for web pages (url) and web search (search).
- taskwarrior integration: tasks persist externally with dependencies and pipeline stages. Multiple agents across sessions reference the same tasks.
- flicknote for plans: persistent, tree-based structure with section IDs. Reviewers target specific sections surgically. Multi-round plan review with 5 specialized subagents (gaps, code, tests, security, docs).
- tmux spawn: each worker gets its own tmux session. No Agent nesting limit. Orchestrator can spawn workers that spawn reviewers that spawn sub-reviewers.
- temenos: OS-native sandbox (seatbelt on macOS, bwrap on Linux). Filesystem allowlists, read-only or write access per agent. MCP server mode under active development.
ttal, organon, temenos, and logos are open source: https://github.com/tta-lab
Full post: https://dev.flicknote.app/notes/1aa741e9-9f32-41e9-8e26-c8b1fdc6d04f
0 comments
—
We Replaced Every Tool Claude Code Ships With · HackerTrans
The problems we hit:
1) No cross-repo exploration. Want an agent to read another project's code? You need to manually configure permissions. There's no "go explore this OSS repo and answer my question."
2) Summarized web fetching. WebFetch is actually a subagent that summarizes a single page into a haiku-length response. Can't trace links or browse related pages. Fetches fresh every time — no caching.
3) Text-level editing. Edit has fuzzy matching, which helps — but it's still raw text. When tree-sitter can give you an AST with named symbols, why make the model reproduce strings to target a function?
4) Ephemeral tasks and plans. Tasks don't persist outside the session. Plans vanish when context resets. Neither supports multi-round review or structured editing.
5) Flat delegation. The Agent tool can't nest — an agent spawned by Agent can't spawn its own sub-agents. One level of delegation, period. A review leader can't spawn specialized sub-reviewers.
6) No isolation. Bash runs on your host. No sandboxing, no filesystem allowlists. You either yolo it or do annoying permission work.
What we built instead:
- ttal ask: multi-mode research (--url caches pages locally with 1-day TTL and follows links, --repo auto-clones OSS repos, --project spawns read-only sandboxed agents in the right directory). Runs on logos, a pure-bash agent loop — works with any LLM provider, so you can use cheaper models for exploration.
- organon (src/url/search): tree-sitter AST editing. See a symbol tree, pick a 2-char ID, pipe new code via heredoc. No text matching. Same pattern for web pages (url) and web search (search).
- taskwarrior integration: tasks persist externally with dependencies and pipeline stages. Multiple agents across sessions reference the same tasks.
- flicknote for plans: persistent, tree-based structure with section IDs. Reviewers target specific sections surgically. Multi-round plan review with 5 specialized subagents (gaps, code, tests, security, docs).
- tmux spawn: each worker gets its own tmux session. No Agent nesting limit. Orchestrator can spawn workers that spawn reviewers that spawn sub-reviewers.
- temenos: OS-native sandbox (seatbelt on macOS, bwrap on Linux). Filesystem allowlists, read-only or write access per agent. MCP server mode under active development.
ttal, organon, temenos, and logos are open source: https://github.com/tta-lab
Full post: https://dev.flicknote.app/notes/1aa741e9-9f32-41e9-8e26-c8b1fdc6d04f