Today I finished most of the work on a set of "AI improvements" for AppWeaver.
1. Intervention mode for agent tool calls
Agents use tools to read and search files, inspect git status, edit code, run commands, and so on.
With intervention mode enabled, you can:
- interrupt/cancel the command before execution.
- add a rule to always allow a tool call, using argument patterns such as `Read **/*.md` to allow reading only Markdown files.
- edit command outputs or interrupt before it's sent back to the agent.
This is especially useful when a command produces huge amount of irrelevant output data that would otherwise consume a lot of context/tokens.
2. View and override system instructions
Most AI tools inject their own system prompts before any user prompt. I made it all transparent and editable by the user (if needed).
3. Skills Manager
Every app in AppWeaver can create skills. But not every time you need all those skills to be enabled. Skills manager lets you enable/disable them at the runtime.
4. FIX: The AGENTS.md file used for developing AppWeaver itself is no longer added to the AI context when the agent is working in a parent workspace (default use-case).
--
AppWeaver gives you more visibility and control over your agents, while helping to reduce unnecessary context/token usage.
dhalsim
npub1de…05uc4
2026-08-13 17:10:51 UTC
Author Public Key
npub1dejts0qlva8mqzjlrxqkc2tmvs2t7elszky5upxaf3jha9qs9m5q605uc4Seen on
wss://nostr.momPublished at
2026-08-13 17:10:51 UTCEvent JSON
{
"id": "1e6b457238ff5767b647f1aab8b3bde568fb5c003f8cac8ae05a9ce7223ece63",
"pubkey": "6e64b83c1f674fb00a5f19816c297b6414bf67f015894e04dd4c657e94102ee8",
"created_at": 1786641051,
"kind": 1,
"tags": [],
"content": "Today I finished most of the work on a set of \"AI improvements\" for AppWeaver.\n\n1. Intervention mode for agent tool calls\n\nAgents use tools to read and search files, inspect git status, edit code, run commands, and so on.\n\nWith intervention mode enabled, you can:\n\n- interrupt/cancel the command before execution.\n- add a rule to always allow a tool call, using argument patterns such as `Read **/*.md` to allow reading only Markdown files.\n- edit command outputs or interrupt before it's sent back to the agent.\n\nThis is especially useful when a command produces huge amount of irrelevant output data that would otherwise consume a lot of context/tokens.\n\n2. View and override system instructions\n\nMost AI tools inject their own system prompts before any user prompt. I made it all transparent and editable by the user (if needed). \n\n3. Skills Manager\n\nEvery app in AppWeaver can create skills. But not every time you need all those skills to be enabled. Skills manager lets you enable/disable them at the runtime. \n\n4. FIX: The AGENTS.md file used for developing AppWeaver itself is no longer added to the AI context when the agent is working in a parent workspace (default use-case).\n\n--\n\nAppWeaver gives you more visibility and control over your agents, while helping to reduce unnecessary context/token usage.",
"sig": "d20292612b6ce0295ee99ba508dbf4fe4349e5e0b81cb9a754c77aa9b23f276a4c44103d88224102f1014c4be4e5b8c98a44359f77aeb40084d590c5fbc7e42f"
}