Cheatsheet

Quick reference for mini-a. Bookmark this page.

Installation

opack install mini-a

Alias (Optional)

# Zsh
echo 'alias mini-a="opack exec mini-a"' >> ~/.zshrc && . ~/.zshrc

If alias setup is not available, run commands as opack exec mini-a [...].

Model Setup

Provider	Config
OpenAI	export OAF_MODEL="(type: openai, model: gpt-5-mini, key: '...')"
Google	export OAF_MODEL="(type: gemini, model: gemini-2.0-flash, key: '...')"
Anthropic	export OAF_MODEL="(type: anthropic, model: claude-sonnet-4-20250514, key: '...')"
Ollama	export OAF_MODEL="(type: ollama, model: 'llama3', url: 'http://localhost:11434')"
Bedrock	export OAF_MODEL="(type: bedrock, options: (region: eu-west-1, model: 'anthropic.claude-sonnet-4-20250514-v1:0'))"
GitHub	export OAF_MODEL="(type: openai, url: 'https://models.github.ai/inference', model: openai/gpt-5, key: $(gh auth token), apiVersion: '')"

Running mini-a

Command	Description
mini-a	Interactive console
mini-a goal='...'	Direct goal
mini-a onport=8080	Web UI
mini-a chatbotmode=true	Chatbot mode
mini-a modelman=true	Model manager
mini-a --skills	Starter self-contained SKILL.yaml template

Console Commands

Command	Action
/help	Show commands
/model [main\|lc\|val]	Choose a model definition for a slot (/model opens an interactive slot picker)
/models	Show all configured model tiers (main, LC, validation) with provider and source
/show [prefix]	Show active parameters
/skills [prefix]	List discovered skills
/compact [n]	Compact older history, keep up to latest n exchanges (default 6)
/summarize [n]	Summarize older history, keep up to latest n exchanges (default 6)
/context	Show token/context breakdown
/reset	Reset conversation
/last [md]	Reprint last final answer (md for raw markdown)
/save <path>	Save last final answer to a file
/stats [mode] [out=file.json]	Usage statistics (summary, detailed, tools, memory, wiki)
/wiki context	Compact wiki overview: page count, sections, mounts, recent log
/wiki list [prefix]	List wiki pages (optionally filtered by prefix); add --meta for title+description
/wiki read <page.md>	Print a wiki page
/wiki search <query>	Full-text search across wiki pages and mounts
/wiki browse [path]	Navigate section structure
/wiki tree [path]	Show full folder hierarchy
/wiki backlinks <page.md>	Show pages linking to this page
/wiki lint	Run wiki health checks (broken links, orphans, stale pages)
/wiki reindex	Rebuild wiki search index (requires wikiaccess=rw)
/wiki move <from.md> <to.md>	Move page and rewrite links (requires wikiaccess=rw)
/wiki init [section/]	Create section index (requires wikiaccess=rw)
/wiki mounts	List active read-only mounts
/wiki attach <name> [backend=fs] [root=path]	Mount a read-only wiki
/wiki detach <name>	Unmount a wiki
/rewind [n]	Undo the last n user+assistant exchanges (default 1); cancels any active subtasks
/dream [memory\|wiki] [plan\|apply\|reorg\|dryrun]	Run memory and/or wiki dream consolidation pass (shown when memorych or usewiki=true is set)
/history [n]	Show recent user goals from conversation history
/exit	Exit mini-a
/clear	Reset conversation history and accumulated metrics
/cls	Clear screen

Key Parameters

Parameter	Default	Description
agent	-	Agent profile (markdown file path or inline markdown)
agentfile	-	Backward-compatible alias for agent
goal	-	The task to accomplish
useshell	false	Enable shell commands
readwrite	false	Allow file writes
chatbotmode	false	Chat-only mode
maxsteps	15	Max agent steps
maxcontext	-	Max context tokens
maxcontent	-	Alias for maxcontext
maxtokens	-	Max output tokens
deepresearch	false	Enable iterative research/validation cycles
validationgoal	-	Quality criteria for deep research
valgoal	-	Alias for validationgoal
vmodel	-	Optional dedicated validation model used in deep-research scoring
modelval	-	Runtime override for the validation model (OAF_VAL_MODEL format)
maxcycles	3	Max deep research cycles
validationthreshold	PASS	Validation verdict/score required to stop
persistlearnings	true	Carry learnings forward between cycles
useplanning	false	Enable planning
useutils	false	Built-in utilities; with usestdutils=true (default) exposes read, glob, grep, webfetch, etc.
usestdutils	true	Expose standard tool aliases instead of legacy Mini Utils names (requires useutils=true)
mini-a-docs	false	Docs-aware Mini Utils root (markdownFiles) when utilsroot is unset
miniadocs	false	Alias for mini-a-docs
useskills	false	Expose skill operations in Mini Utils Tool (requires useutils=true)
skillmaxautoload	1	Maximum number of high-confidence matching skills to auto-load into bounded runtime context
skillcontextchars	8000	Maximum characters read from each auto-loaded SKILL.md for runtime context
skillmanifestchars	1536	Approximate character budget for skill descriptions in the system prompt manifest
usetools	false	Enable tool use
usetoolslc	false	Register MCP tools only on the low-cost model; main model stays in prompt/action mode
usejsontool	false	Compatibility json tool for some tool-calling models
toollog	-	JSSLON channel for MCP tool call logs (input/output)
auditch	-	Channel for agent activity audit log (tool calls, shell commands, goal events)
debugch	-	Channel for main-model LLM request/response payloads
debuglcch	-	Channel for low-cost model LLM request/response payloads
debugfile	-	Write debug output to a NDJSON file (implies debug=true)
debugvalch	-	Separate debug channel for the validation model
outfileall	-	Deep research only: save full cycle outputs (not only final answer)
shelltimeout	-	Max shell command runtime (ms) before timeout
shellmaxbytes	8000	Cap shell output size and truncate with head/tail excerpt
shellallowpipes	false	Allow pipes, redirection, and shell control operators
usestream	false	Stream responses
showexecs	false	Show shell/exec events as separate lines in the interaction stream
showseparator	true	Show a separator line between interaction events (disable for compact view)
browsercontext	-	Browser context configuration or true to auto-enable browser control for supported tools
showthinking	false	Surface XML-tagged <thinking>...</thinking> blocks as thought logs
promptprofile	context-dependent	System prompt verbosity: minimal, balanced, or verbose (minimal in chatbot mode; verbose with debug=true; otherwise balanced)
systempromptbudget	—	Max estimated tokens for the system prompt; drops lower-priority sections when exceeded
goalprefix	—	Prefix automatically prepended to every goal before the agent sees it
compressgoal	false	Enable automatic compression of oversized goal text before execution
compressgoaltokens	250	Estimated token threshold before goal compression is considered
compressgoalchars	1000	Character threshold before goal compression is considered
usemath	false	Enable LaTeX math guidance for KaTeX rendering in web UI
usesvg	false	Enable SVG generation for custom visuals and infographics
usediagrams	false	Enable diagram generation
usemermaid	false	Alias for usediagrams
usecharts	false	Enable chart generation
usevectors	false	Enable vector bundle (usesvg=true + usediagrams=true)
mcp	-	MCP servers to load
mcpproxy	false	MCP proxy mode
mcpproxythreshold	0	Proxy auto-spill byte threshold (0 disables); large results written to temp file
mcpproxytoon	false	TOON serialization for spilled proxy results
mcpproxyallow	-	Comma-separated allowlist of downstream tool names exposed through proxy-dispatch
mcpproxydeny	-	Comma-separated denylist of downstream tool names hidden from proxy-dispatch (applied after mcpproxyallow)
mcpprogcall	false	Enable localhost bridge for programmatic MCP tool calls
mcpprogcallport	0	Programmatic MCP bridge port (0 auto-selects)
mcpprogcallmaxbytes	4096	Max inline bridge response size before spill
mcpprogcallresultttl	600	TTL (seconds) for spilled bridge results
mcpprogcallbatchmax	10	Max tool calls accepted per bridge batch request
maxpromptchars	120000	Max accepted prompt size for incoming web prompts
lccontextlimit	0	Escalate to main model when low-cost model context gets too large
deescalate	3	Successful steps before returning from main model to low-cost model
modellock	auto	Force model tier: main, lc, or auto
modelstrategy	default	Model orchestration profile: default (adaptive LC-first), advisor (LC executor + main as selective advisor), or delegate (LC runs all steps including step 0)
advisorenable	true	Enable advisor consultations when modelstrategy=advisor
advisormaxuses	2	Max advisor consultations per run when modelstrategy=advisor
advisoronrisk	true	Allow advisor consults on risk signals
advisoronambiguity	true	Allow advisor consults on ambiguity signals
advisoronharddecision	true	Allow advisor consults for hard-decision checkpoints
advisorcooldownsteps	2	Min steps between advisor consultations
advisorbudgetratio	0.20	Fraction of session token budget advisor calls may consume
emergencyreserve	0.10	Portion of advisor budget reserved for high-risk/high-value consults
harddecision	warn	Hard-decision checkpoint mode: require, warn, or off
evidencegate	false	Enable lightweight evidence gating for non-trivial actions
evidencegatestrictness	medium	Tuning level for evidence gate heuristics: low, medium, or high
nologtrunc	false	Disable truncation of long log output lines (show full content)
usewiki	false	Enable wiki knowledge base
wikiaccess	ro	Wiki access: ro (read-only) or rw (read-write)
wikibackend	fs	Wiki backend: fs (filesystem), s3, s3fs, or es (Elasticsearch/OpenSearch)
wikiroot	.	Root directory for the fs backend
wikibucket	-	S3 bucket name (s3 backend)
wikimounts	-	Read-only wiki mounts (SLON/JSON array): [{name: 'team', backend: 'fs', root: '/path'}]
usememory	false	Enable the working memory subsystem
memoryuser	false	Convenience preset: file-backed global + session memory at ~/.openaf-mini-a/, auto-promote facts,decisions,summaries, 30-day stale sweep
memoryusersession	false	Convenience preset: enables usememory, session-only file-backed channel under ~/.openaf-mini-a/
memoryscope	both	Which store the agent reads from/writes to: session, global, or both
memorych	-	SLON/JSON definition of an OpenAF channel to persist the global memory store
memorysessionch	-	SLON/JSON definition of a channel for session memory persistence (falls back to memorych)
memorysessionid	<agent-id>	Session key used to namespace session memory in the channel
memorymaxpersection	80	Maximum entries kept per section before compaction drops stale/old ones
memorymaxentries	500	Hard cap on total entries across all sections
memorycompactevery	8	How many appends trigger an automatic compaction pass
memorydedup	true	Suppress near-duplicate entries (85% word-overlap fingerprint)
memorypromote	""	Auto-promote selected sections from session to global memory at session end
memorystaledays	0	Mark unconfirmed global memory entries as stale after N days
memoryinject	summary	Inject memory as section counts (summary with dynamic search) or full compact entries (full)
memorysessionheader	—	HTTP request header name used to derive the memory session ID in web mode (e.g. X-User-Id)
llmcomplexity	false	Validate medium-complexity routing with an LC model check
metricsch	-	SLON/JSON channel for recording periodic metrics snapshots — see Channels (note: backend options nest under options:, e.g. (name: 'mini-a-metrics', type: 'mvs', options: (file: 'metrics.db'), period: 5000))
usedelegation	false	Agent delegation
workers	-	Remote worker URLs for delegation
workertags	—	Comma-separated tags appended to the default worker skill in the AgentCard
usea2a	false	Use A2A HTTP+JSON endpoints for remote worker delegation
apitoken	—	Bearer token for authenticating requests to the worker API server
autodelegation	false	Auto-delegate noisy/oversized tool results to a summarization sub-agent (requires usedelegation=true)
autodelegationthreshold	8192	Byte size threshold that triggers auto-delegation
autodelegationmaxperstep	2	Max auto-delegations per agent step
noisytools	""	Comma-separated tool names always auto-delegated regardless of result size
subtasks	""	Pipe-separated startup scout goals submitted before the main loop
subtasksfile	""	Path to YAML/JSON file of {goal, fork, args, timeout} startup task objects
subtaskssequential	false	Run startup scouts (and all subtasks) one at a time instead of in parallel
forkstatemaxbytes	65536	Max bytes of serialized fork state transmitted to a remote worker
toolfallback	false	Fall back to action mode when malformed pseudo tool calls are emitted by the model
dream	false	Run in standalone dream-pass mode (memory/wiki consolidation) instead of a regular session
dryrun	false	Preview dream-pass changes without writing anything back
dreamwikimode	apply	Wiki dream mode: lint, plan, apply, reorg
dreammemorymode	apply	Memory dream mode: plan or apply
dreamwikiapply	false	Required write gate for wiki apply/reorg
dreamwikiapproval	ask	Reorg approval mode: auto, ask, never
dreamwikireorg	false	Allow structural wiki reorg
dreamreport	—	Optional JSON output report path
dreammaxsteps	60	Maximum agent steps for the wiki dream pass
maxauditrecords	200	Maximum audit log entries included in the memory dream consolidation prompt
onport	-	Web UI port
maxpromptchars	120000	Max accepted prompt size for incoming web prompts
ssequeuetimeout	120	Web SSE stream queue timeout in seconds
logpromptheaders	-	Comma-separated HTTP request header names to log alongside incoming web prompts
usehistory	false	Enable conversation history persistence in web mode
useattach	false	Enable file attachment support in web mode
historykeep	false	Keep finished web conversation history files instead of discarding them
historypath	-	Directory path used to store web conversation history files
historyretention	600	Web history retention window in seconds
historykeepperiod	-	Delete kept conversation files older than this many minutes
historykeepcount	-	Keep only the newest N kept conversation files
historys3bucket	-	S3 bucket used to mirror history files
historys3prefix	-	S3 key prefix for mirrored history files
historys3url	-	S3 endpoint URL for history mirroring
historys3accesskey	-	S3 access key for history mirroring
historys3secret	-	S3 secret key for history mirroring
historys3region	-	S3 region for history mirroring
historys3useversion1	false	Use S3 path-style (v1) signing for history mirroring
historys3ignorecertcheck	false	Disable TLS certificate checks for history S3 access
homedir	—	Override the home directory used to locate the .openaf-mini-a folder

Agent Files

Load a complete agent profile from a markdown file with YAML frontmatter:

mini-a agent=my-agent.agent.md goal="..."
mini-a --agent   # Print a starter template

Supported frontmatter keys: name, description, model, capabilities, tools, constraints, rules, knowledge, youare, mini-a (parameter overrides).

agentfile= is a backward-compatible alias for agent=.

File Inclusion

mini-a useutils=true goal='@file.txt Summarize this file'
mini-a useutils=true goal='@data.csv Analyze it'

Custom Commands, Skills, Hooks

Item	Path/Command
Custom slash template	~/.openaf-mini-a/commands/<name>.md
Skill (folder)	~/.openaf-mini-a/skills/<name>/SKILL.md
Skill (file)	~/.openaf-mini-a/skills/<name>.md
Hooks	~/.openaf-mini-a/hooks/*.yaml|*.yml|*.json
Extra command dirs	extracommands=/path/a,/path/b
Extra skill dirs	extraskills=/path/a,/path/b
Extra hook dirs	extrahooks=/path/a,/path/b
Alternate config base	homedir=/opt/shared/mini-a-config
Run one template	mini-a exec="/<name> arg1 arg2"
List skills	/skills

Mode Presets

Mode	Enables
shell	Read-only shell access (useshell=true)
shellrw	Shell + write access + non-interactive shell approvals
utils	Mini Utils Tool preset (useutils=true mini-a-docs=true usetools=true)
chatbot	Chat-only mode
internet	Internet-focused MCP/tool mode with docs-aware utils
news	Internet + RSS news-focused MCP mode
poweruser	High-capability preset with shell, utils, proxy tuning, and docs-aware defaults
web	Browser UI preset with MCP tools enabled
webfull	Full web UI preset with history, attachments, proxying, and richer rendering modes

Custom modes: create ~/.openaf-mini-a_modes.yaml with a modes: map. Custom definitions are merged with built-ins and override duplicates.

# ~/.openaf-mini-a_modes.yaml
modes:
  mypreset:
    useshell: true
    readwrite: true
    maxsteps: 30

mini-a mode=mypreset goal='your goal here'

Dual-Model Setup

export OAF_MODEL="(type: openai, model: gpt-5-mini, key: '...')"
export OAF_LC_MODEL="(type: openai, model: gpt-5-nano, key: '...')"
export OAF_VAL_MODEL="(type: openai, model: gpt-5-mini, key: '...')"
# Saves 50-70% on token costs

# Optional prompt-mode controls (Gemini main auto-enables when unset)
export OAF_MINI_A_NOJSONPROMPT=true
export OAF_MINI_A_LCNOJSONPROMPT=true

Full Debug

Capture everything — agent activity, main-model payloads, and low-cost model payloads — to separate JSON files:

mini-a goal="your goal" \
  auditch="(type: file, options: (file: audit.json))" \
  debugch="(type: file, options: (file: debug.json))" \
  debuglcch="(type: file, options: (file: debuglc.json))"

Add debugvalch when using a validation model (deepresearch=true):

mini-a goal="research task" deepresearch=true \
  auditch="(type: file, options: (file: audit.json))" \
  debugch="(type: file, options: (file: debug.json))" \
  debuglcch="(type: file, options: (file: debuglc.json))" \
  debugvalch="(type: file, options: (file: debugval.json))"

Config Home (homedir)

By default Mini-A reads ~/.openaf-mini-a. Override with homedir=:

# Shared team config
mini-a homedir=/opt/shared/mini-a-config goal="..."

# Per-project isolated config
mini-a homedir=./my-project-config goal="..."

# Shared base + extra project skills
mini-a homedir=/opt/shared/mini-a-config extraskills=./project-skills goal="..."

Docker

# Interactive
docker run -it -e OAF_MODEL="(type: openai, model: gpt-5-mini, key: '...')" openaf/mini-a
# Web UI
docker run -p 8080:8080 -e OAF_MODEL="(type: openai, model: gpt-5-mini, key: '...')" openaf/mini-a onport=8080

Common MCP Servers

Server	Tools	Example
mcp-time	Current time, timezone conversion	mcp="(cmd: 'ojob mcps/mcp-time.yaml')"
mcp-db	SQL queries, schema inspection	mcp="(cmd: 'ojob mcps/mcp-db.yaml')"
mcp-web	Web search, fetch pages	mcp="(cmd: 'ojob mcps/mcp-web.yaml')"
mcp-file	File operations	mcp="(cmd: 'ojob mcps/mcp-file.yaml')"
mcp-shell	Shell command execution	mcp="(cmd: 'ojob mcps/mcp-shell.yaml')"