@@ -9,8 +9,6 @@ A coding agent via a augmented shell prompt, small and fast that doesn't get in
99 - Codex: https://github.com/openai/codex Powerful and open source, focused on heavy open-ai integration
1010 - Opencode: https://opencode.ai/ The open source standard. Similar to oh-my-pi but with a different aesthetic
1111
12-We want to do our best to support their configs, try to find the common api and auto load them along with the community .agents conventions.
13-
1412 ## Philosophy:
1513
1614 **Fast and performant**, instant feedback shell like output. Performance first style second.
@@ -19,39 +17,54 @@ Very focused on dev flow, simple setup steps and fast to coding.
1917 Elegant output that focus on the conversation with the agent and their actions. Ensure an accurate cronological conversation log that is very readable for the user. Keep a visible hierchy of actions -> results and assistant reasoning and responses.
2018 Accurate and reliable and fast core set of features.
2119 Community oriented, all of the inspirations have their own dotfile format, try to follow the existing
22-conventions and not introduce more specs. (`.agents` / `AGENTS.md`, while also supporting adjacent conventions like `.claude`)
20+conventions and not introduce more specs. We want to do our best to support and autodiscover community config standards:
21+
22+- AGENTS.md -> https://agents.md/
23+ - CLAUDE.md (same as AGENTS.md)
24+- Skills -> https://agentskills.io/client-implementation/adding-skills-support
25+ - Claude skills too. They claim to follow the spec, we should only need to discover them too.
2326
2427 ## Features:
2528
2629 - Auto discovery of providers via ENV (Example: OPENCODE_API_KEY), or local servers (Example: ollama)
30+ - `/login` shows OAuth login status. `/login <provider>` starts browser-based OAuth login. `/logout <provider>` clears saved tokens. Currently supports Anthropic. (Both pi and opencode support this too)
31+ - `/model` (alias `/models`) command allows the user to pick a model from connected providers. As well as thinking effort for the model if supported. Selection persists accross sessions. Must have autocomplete support for ease of use when changing models
32+
2733 - Session management and command to create new/resume/list with local sqlite file.
28-- Rich line editing/history/image paste
29-- Seamless shell integration with `!` in prompt input.
30-- Reference files and skills from the working directory and global configs with `@` in prompt input, plus autocomplete for files, skills, to include in the prompt. Skills should have their respective tools and implementation to match other coding agents
31-- Press `ESC` at any point during an assistant response to interrupt it: the partial response is preserved in history with an interrupt stub appended (so the LLM retains context), and the user is returned to the prompt silently. `ctrl+c` exits forcefully. `ctrl+d` (EOF) gracefully exits.
32-- Elegantly handles hitting max context size for the model using regular pruning.
33-- No max steps or max tool calls. user can interrupt, this is not needed.
34-- Support prompt caching via the used sdks.
35-- Feature parity with community configs like custom agents, skill and commands.
36-- Shell-first tool surface for LLMs: `shell`, `listSkills`, and `readSkill`, plus connected MCP tools and optional web tools when configured. Keep the core surface small and easy to use so it actually reduces friction for the LLMs.
37-- File work should follow a simple flow: inspect with shell, mutate with `mc-edit`, verify with shell.
38-- Large shell outputs should be truncated automatically for the llm, to avoid context explosions and encourage targetted reads.
39-- `mc-edit` should stay narrow and reliable: exact-text edits only, deterministic failures on stale or ambiguous state, diff and machine-friendly output while staying human readable.
40-- Optional `webSearch` and `webContent` tools when `EXA_API_KEY` is set.
41-- Commands in CLI prompt:
42- - `/model` (alias `/models`) command allows the user to pick a model from connected providers. As well as thinking effort for the model if supported. Selection persists accross sessions.
34+ - `/session` command to interactively manage sessions as well as cli flags.
35+ - `/new` starts a new session with clean context. Clean UI and fresh session display
36+
37+- Enriched user prompt:
38+ - Rich line editing/history/image paste
39+ - Seamless shell integration with `!` in prompt input. Runs command in subprocess and sends the output to the llm as user message.
40+ - Reference skills from the working directory and global configs with `/` in prompt input, plus autocomplete for skills to include in the prompt.
41+ - Press `ESC` at any point during an assistant response to interrupt it: the partial response is preserved in history with an interrupt stub appended (so the LLM retains context), and the user is returned to the prompt silently. `ctrl+c` exits forcefully. `ctrl+d` (EOF) gracefully exits.
42+ - Image support in prompt input, including pasted image data URLs and pasted image file paths.
43+ - Shell like autocomplete for filepaths, `@` embeds a file into the prompt and autocompletes filepaths
44+
45+- Context handling:
46+ - Elegantly handles hitting max context size for the model using regular pruning.
47+ - No max steps or max tool calls. user can interrupt, this is not needed.
48+ - Support prompt caching via the used sdks.
49+ - Rolling context pruning with a balanced default. (Must not break caching!)
50+ - Recover from errors returning the user to the prompt with clear messaging
4351 - `/undo` removes the last turn from conversation history, it does not restore filesystem state.
52+
53+- Shell-first tool surface for LLMs: `shell`, `listSkills`, and `readSkill`, plus connected MCP tools and optional web tools when configured. Keep the core surface small and easy to use so it actually reduces friction for the LLMs.
54+ - File work should follow a simple flow: inspect with shell, mutate with `mc-edit`, verify with shell.
55+ - Large shell outputs should be truncated automatically for the llm, to avoid context explosions and encourage targetted reads.
56+ - `mc-edit` should stay narrow and reliable: exact-text edits only, deterministic failures on stale or ambiguous state, diff and machine-friendly output while staying human readable.
57+ - Optional `webSearch` and `webContent` tools when `EXA_API_KEY` is set.
58+ - Cross model system prompt with clear descriptions of the expected dev flow and guidance as a code agent focused on shell first.
59+
60+-Other Commands in CLI prompt:
4461 - `/reasoning` toggles display of model reasoning output.
4562 - `/verbose` toggles output truncation. Tuncates large outputs keeping start and end sections and truncating the rest when verbose is off. This controls truncation for the user output, it's a visibility UI concern
46- - `/context` configures context pruning and tool-result caps.
47- - `/review` reviews recent changes via a global custom command installed at app start (`~/.agents/commands/review.md`), and can be customized or shadowed locally.
48- - `/agent` sets or clears the active agent.
49- - `/mcp` list/add/remove mcp servers. servers are stored in sqlite
50- - `/login` shows OAuth login status. `/login <provider>` starts browser-based OAuth login. `/logout <provider>` clears saved tokens. Currently supports Anthropic.
51- - `/new` starts a new session with clean context. Clean UI and fresh session display
63+ - `/review` reviews recent changes via a global custom skill installed at app start (`~/.agents/skills/review.md`), and can be customized or shadowed locally.
64+
5265 - Connect to MCP servers over Streamable HTTP / SSE fallback or stdio.
53-- Image support in prompt input, including pasted image data URLs and pasted image file paths.
54-- Excellent auto complete for all commands, subtituitions. The user should also be able to reference files with autocomplete wihtout including them into the prompt like `@` does with TAB
66+ - `/mcp` list/add/remove mcp servers. servers are stored in sqlite
67+
5568
5669 ## UI/Output