xonecas.io
64875
Mc

mini-coder / commit 8969dbf

refactor: combine mini-coder and orchestrator into a single skill

Merge mini-coder-orchestrator/SKILL.md content into mini-coder/SKILL.md
as an 'Orchestration mode' section. Remove the separate orchestrator
skill directory. The combined skill follows the Agent Skills spec with
proper frontmatter, progressive disclosure, and stays under 500 lines.

xonecas2026-03-20 8969dbf59113a16fce9d04f0e3f77a5ff8d8fd7c parent 59cfedd

2 files changed

.agents/skills/mini-coder-orchestrator/SKILL.md+0−35
@@ -1,35 +0,0 @@
1----
2-name: orc
3-description: Orchestrator mode - decomposes complex tasks and coordinates a fleet of subagents
4----
5-
6-You are in **Orchestration** mode, behave like an expert orchestration agent. Your superpower is breaking complex problems into well-scoped subtasks and delegating each via the `mc` shell tool to other agents.
7-
8-## Core principles
9-
10-- **Decompose first.** Before using any tool, think about the full shape of the task. Identify subtasks that are independent and self-contained — those are candidates for delegation.
11-- **Delegate aggressively.** Prefer spawning a focused subagent over doing broad work yourself. Each subagent gets a single, precise goal.
12-- **Stay the coordinator.** You own the big picture. Subagents handle the details. Synthesise their results, resolve conflicts, and report back clearly.
13-- **Parallelise when possible.** If two subtasks don't depend on each other, dispatch them concurrently.
14-- **Fail fast.** If a subagent reports failure or returns unexpected output, stop and reassess before continuing.
15-- **Custom agents**: Use these if appropriate, otherwise a normal agent is preffered.
16-
17-## How to orchestrate
18-
19-1. Read the user's goal carefully.
20-2. Identify the 2–5 discrete subtasks needed to complete it.
21-3. Write a tight, self-contained prompt for each — include all context the agent needs; it has no shared memory with you.
22-4. Dispatch agents (in parallel where safe) and capture their outputs.
23-5. Synthesise results. If something is missing or wrong, spawn a follow-up agent to fix it.
24-6. Report the final outcome clearly to the user.
25-
26-## Subagent prompt hygiene
27-
28-- Be explicit about _what the agent should produce_ (a file, a code change, a summary, a decision).
29-- Include relevant paths, constraints, and acceptance criteria in every prompt.
30-- Never give a vague brief — ambiguity leads to rework and wasted tokens.
31-
32-## When NOT to delegate
33-
34-- Trivial lookups or single-file reads: do them directly with your own tools.
35-- Tasks that require continuous back-and-forth with the user: handle interactively.
.agents/skills/mini-coder/SKILL.md+72−156
@@ -1,207 +1,123 @@
11 ---
22 name: mini-coder
3-description: How to use the `mc` CLI coding agent
3+description: How to use the `mc` CLI coding agent — interactive sessions, one-shot prompts, slash commands, file editing with mc-edit, and orchestrating subagents for complex tasks.
44 ---
55
66 # mini-coder
77
8-Use this skill when you need to help a user operate the `mc` command-line tool.
8+Use this skill when you need to help a user operate the `mc` command-line tool, or when you are acting as an `mc` agent yourself.
99
10-`mc` is the executable for **mini-coder**, a small CLI coding agent. This skill is about **how to use the tool**, not how the mini-coder repo is implemented.
10+`mc` is the executable for **mini-coder**, a small CLI coding agent. This skill covers **how to use the tool**, not how the mini-coder repo is implemented.
1111
12-## What `mc` does
13-
14-- Starts an interactive coding session in the terminal
15-- Accepts a one-shot prompt from the command line
16-- Persists sessions locally so they can be resumed
17-- Lets users switch models, manage MCP servers, and load custom agents/skills/commands
18-
19-## Core command-line usage
20-
21-```sh
22-mc
23-mc "explain this codebase"
24-mc -c
25-mc -r <session-id>
26-mc -l
27-mc -m <model-id>
28-mc --cwd <path>
29-mc -h
30-```
31-
32-## Most common flows
33-
34-### Start an interactive session
35-
36-```sh
37-mc
38-```
39-
40-Use this when the user wants a normal back-and-forth coding session.
41-
42-### Run a prompt immediately
12+## Command-line usage
4313
4414 ```sh
45-mc "refactor the auth module to use async/await"
15+mc # interactive session
16+mc "explain this codebase" # one-shot prompt
17+mc -c # continue most recent session
18+mc -r <session-id> # resume a specific session
19+mc -l # list recent sessions
20+mc -m <model-id> # pick a model
21+mc --cwd <path> # run in another directory
22+mc -h # help
4623 ```
4724
48-Use this when the user already knows the task they want to give `mc`.
49-
50-### Continue a previous session
51-
52-```sh
53-mc -c
54-mc -r <session-id>
55-mc -l
56-```
57-
58-- `mc -c` continues the most recent session
59-- `mc -r <session-id>` resumes a specific session
60-- `mc -l` lists recent sessions so the user can find an ID
61-
62-### Pick a model
63-
64-```sh
65-mc -m zen/claude-sonnet-4-6
66-mc -m ollama/llama3.2
67-```
68-
69-Use `-m` when the user wants a specific provider or local model.
70-
71-### Run in another directory
72-
73-```sh
74-mc --cwd ~/src/other-project
75-```
76-
77-Useful when the user is not already inside the target repo.
78-
7925 ## Interactive slash commands
8026
81-Inside `mc`, the user can run slash commands such as:
82-
83-- `/help` — show help
84-- `/model` — list models
85-- `/model <id>` — switch models
86-- `/model effort <low|medium|high|xhigh|off>` — set reasoning effort when supported
87-- `/reasoning [on|off]` — toggle reasoning display
88-- `/undo` — remove the last conversation turn (does not revert filesystem changes)
89-- `/new` — start a fresh session
90-- `/mcp list` — list configured MCP servers
91-- `/mcp add <name> http <url>` — add an HTTP MCP server
92-- `/mcp add <name> stdio <cmd> [args...]` — add a stdio MCP server
93-- `/mcp remove <name>` — remove an MCP server
94-- `/agent [name]` — set or clear an active primary custom agent
95-- `/login` — show OAuth login status
96-- `/login <provider>` — login via OAuth (opens browser). Currently supports `anthropic`
97-- `/logout <provider>` — clear saved OAuth tokens
98-- `/exit`, `/quit`, `/q` — leave the session
27+Inside `mc`:
28+
29+| Command | Purpose |
30+| ----------------------------------------------- | --------------------------------------------- |
31+| `/help` | Show help |
32+| `/model` | List models |
33+| `/model <id>` | Switch model |
34+| `/model effort <low\|medium\|high\|xhigh\|off>` | Set reasoning effort |
35+| `/reasoning [on\|off]` | Toggle reasoning display |
36+| `/verbose` | Toggle output truncation |
37+| `/undo` | Remove last turn (does not revert filesystem) |
38+| `/new` | Start a fresh session |
39+| `/review` | Review recent changes |
40+| `/mcp list` | List MCP servers |
41+| `/mcp add <name> http <url>` | Add HTTP MCP server |
42+| `/mcp add <name> stdio <cmd> [args...]` | Add stdio MCP server |
43+| `/mcp remove <name>` | Remove MCP server |
44+| `/agent [name]` | Set or clear active custom agent |
45+| `/login [provider]` | Show or start OAuth login |
46+| `/logout <provider>` | Clear OAuth tokens |
47+| `/exit`, `/quit`, `/q` | Leave the session |
9948
10049 ## Prompt input features
10150
102-### Shell input
103-
104-If the user starts a prompt with `!`, `mc` runs that shell command inline and can use the result in context.
105-
106-mini-coder itself is shell-first: repo inspection happens primarily through shell commands, and targeted edits are typically done via `mc-edit` invoked from shell.
107-
108-### References with `@`
109-
110-`mc` supports `@` references in prompts for:
111-
112-- files
113-- custom agents
114-- skills
115-
116-This helps users quickly pull relevant context into a prompt.
51+- **Shell input** — prefix with `!` to run a shell command inline and feed output to the LLM.
52+- **File references** — `@path/to/file` embeds a file into the prompt (autocompletes paths).
53+- **Skill / agent refs** — `/` triggers autocomplete for skills and agents.
54+- **Images** — paste image data URLs or file paths directly.
55+- **ESC** interrupts an in-progress response (preserves partial output). `Ctrl+C` exits forcefully, `Ctrl+D` exits gracefully.
11756
11857 ## File editing with mc-edit
11958
120-`mc` is shell-first. The preferred workflow for file edits is: inspect with shell → edit with `mc-edit` → verify with shell.
121-
122-`mc-edit` is a helper command invoked from the shell tool for targeted, exact-text file edits:
59+`mc` is shell-first. The preferred workflow is: **inspect → edit → verify**.
12360
12461 ```sh
12562 mc-edit <path> (--old <text> | --old-file <path>) [--new <text> | --new-file <path>] [--cwd <path>]
12663 ```
12764
128-- Applies one exact-text replacement to an existing file
129-- The old text must match exactly once (fails on zero or multiple matches)
130-- Omit `--new`/`--new-file` to delete the matched text
131-- On success: prints a unified diff + metadata (`ok`, `path`, `changed`)
132-- On no-op: prints `(no changes)` + metadata
133-- Errors go to stderr
65+- Applies one exact-text replacement per invocation.
66+- Old text must match exactly once (fails on zero or multiple matches).
67+- Omit `--new` / `--new-file` to delete matched text.
68+- Prints a unified diff + metadata on success; errors go to stderr.
13469
13570 ### Examples
13671
137-Replace a function signature:
138-
13972 ```sh
14073 mc-edit src/auth.ts --old "function login(user)" --new "async function login(user)"
141-```
142-
143-Delete a block:
14474
145-```sh
14675 mc-edit src/config.ts --old "// TODO: remove this
14776 const DEBUG = true;"
148-```
149-
150-Use temp files for large edits:
15177
152-```sh
153-mc-edit src/big-file.ts --old-file /tmp/old.txt --new-file /tmp/new.txt
78+mc-edit src/big.ts --old-file /tmp/old.txt --new-file /tmp/new.txt
15479 ```
15580
15681 ## Config locations
15782
158-mini-coder supports both `.agents` and `.claude` config layouts.
83+mini-coder discovers configs from both `.agents` and `.claude` layouts:
15984
160-### Local config in a repo
85+| Scope | Paths |
86+| ------ | -------------------------------------------------------------------------------------- |
87+| Local | `.agents/commands/*.md`, `.agents/agents/*.md`, `.agents/skills/<name>/SKILL.md` |
88+| Global | `~/.agents/commands/*.md`, `~/.agents/agents/*.md`, `~/.agents/skills/<name>/SKILL.md` |
16189
162-- `.agents/commands/*.md`
163-- `.agents/agents/*.md`
164-- `.agents/skills/<name>/SKILL.md`
90+Same structure applies under `.claude/` and `~/.claude/`.
16591
166-### Global config in the home directory
92+App data (sessions, MCP config, logs) lives in `~/.config/mini-coder/`.
16793
168-- `~/.agents/commands/*.md`
169-- `~/.agents/agents/*.md`
170-- `~/.agents/skills/<name>/SKILL.md`
94+## Orchestration mode
17195
172-Same ideas also work under `.claude/` and `~/.claude/`.
96+For complex, multi-part tasks you can use `mc` as a subagent coordinator.
17397
174-## Data and state
98+### Core principles
17599
176-mini-coder stores app data in:
100+- **Decompose first.** Break the task into 2–5 independent, self-contained subtasks.
101+- **Delegate aggressively.** Spawn focused subagents via `mc "prompt"` in the shell tool — each gets a single precise goal.
102+- **Parallelise when possible.** Dispatch independent subtasks concurrently.
103+- **Stay the coordinator.** You own the big picture; subagents handle the details. Synthesise their results, resolve conflicts, report back clearly.
104+- **Fail fast.** If a subagent fails, stop and reassess before continuing.
177105
178-```text
179-~/.config/mini-coder/
180-```
181-
182-This includes session history, MCP server config, and related local metadata.
183-
184-## How to help users effectively
106+### Subagent prompt hygiene
185107
186-When answering questions about `mc`:
108+- Be explicit about what the agent should produce (a file, a code change, a summary).
109+- Include all relevant paths, constraints, and acceptance criteria — subagents have no shared memory.
110+- Never give vague briefs; ambiguity leads to rework and wasted tokens.
187111
188-- Prefer giving exact commands the user can paste
189-- Distinguish clearly between **shell commands** (run before or around `mc`) and **slash commands** (run inside `mc`)
190-- Suggest `mc -h` or `/help` when the user needs command discovery
191-- Suggest `mc -l`, `mc -c`, or `mc -r <id>` for session recovery
192-- Mention `--cwd` when the problem is about the wrong working directory
193-- Mention `-m <model-id>` when the problem is model selection
194-- Mention MCP commands only when the user is integrating external tools
112+### When NOT to delegate
195113
196-## Examples to suggest
114+- Trivial lookups or single-file reads — do them directly.
115+- Tasks requiring continuous back-and-forth with the user — handle interactively.
197116
198-```sh
199-mc
200-mc "add tests for src/internal/file-edit/exact-text.ts"
201-mc -c
202-mc -l
203-mc -m zen/claude-sonnet-4-6
204-mc --cwd ~/src/my-project
205-```
117+## Tips for helping users
206118
207-If a user asks how to do something in mini-coder, answer in terms of the `mc` CLI and its interactive commands.
119+- Prefer exact commands the user can paste.
120+- Distinguish **shell commands** (outside `mc`) from **slash commands** (inside `mc`).
121+- Suggest `mc -h` or `/help` for discovery.
122+- Suggest `mc -l` / `mc -c` / `mc -r <id>` for session recovery.
123+- Mention `--cwd` for wrong-directory issues, `-m` for model selection.