Use this page when you want SuperQode to become your main coding-agent harness or sit beside OpenCode, Pi, fast-agent, Codex CLI, Claude Code, Antigravity, and other local tools.
The practical path is:
- Connect the runtime or agent you already use.
- Keep sessions readable, forkable, exportable, and portable.
- Put local plugins and MCP behind project trust.
- Move repeatable work into SuperQode harness specs over time.
superqode --tuiConnection commands:
:connect
:connect codex # Codex SDK using your local Codex login
:runtime claude-agent-sdk # Claude Agent SDK using ANTHROPIC_API_KEY
:connect antigravity # signed-in agy CLI; run `agy` once to log in
:connect acp # any configured ACP agent
:connect byok # direct hosted provider/API key
:connect local # Ollama, LM Studio, MLX, vLLM, SGLang, TGI, DS4
Optional Vim-like terminal navigation:
:vim on
:vim tutor
: # commands
/pattern # transcript search
j / k # conversation and picker navigation
gg / G # transcript beginning or end
:w
:e <file>
:ls
:grep <term>
q:
@:
Vim mode is modal and persistent. It provides Normal, Insert, Command, and Search states for operating SuperQode from the keyboard. See Vim-Like Terminal Navigation.
In the TUI:
:tree
:session
:session rename <name>
:sessions
:sessions rename <id-or-name> <title>
:sessions switch
:sessions switch <id-or-name>
:sessions switch latest
:resume latest
:resume <id-or-name>
:fork <new-id>
:export html
:export markdown
:export json
:sessions lists BYOK and HarnessSpec sessions for this directory grouped by
harness, with model · topic · age rows. :sessions switch restores harness,
provider/model, cwd, and the same transcript. :pipy resume <n> reopens a prior
PiPy file into the live chat and registers it for :sessions.
From the CLI:
superqode sessions list
superqode sessions tree
superqode sessions show <session-id>
superqode sessions export <session-id> --format markdown --output session.md
superqode sessions export <session-id> --format json --output session.jsonSuperQode supports local/offline share artifacts today. They are plain JSON
files using the superqode-share-v1 format.
In the TUI:
:share
:share create [session] [path]
:share export [session] [path] [--json|--markdown]
:share import <artifact.superqode-share.json> [new-session-id]
:share list
:share revoke <artifact>
From the CLI:
superqode share create <session-id>
superqode share export <session-id> --format markdown -o session.md
superqode share export <session-id> --format json -o session.json
superqode share list --json
superqode share import <artifact.superqode-share.json> --session-id imported-session
superqode share revoke <artifact>Use :tree or superqode sessions tree to inspect fork lineage before sharing.
See Session Commands and Share Commands for the full API.
Trust gates plugins and MCP operations. Check status with :trust status or superqode trust status. Grant trust with :trust yes or superqode trust yes. Validate the trust store and risk signals with :trust doctor or superqode trust doctor.
See Trust Commands for details.
Project plugins live under .superqode/plugins/<id>/plugin.json. Install with :plugins add <path> or superqode plugins add <path>. Requires project trust. Validate manifests and references with :plugins doctor or superqode plugins doctor.
See Plugin Commands for the manifest format, hook points, and full CLI reference.
Store project facts and preferences with :memory remember "text" or superqode memory remember "text". Search with :memory search query or superqode memory search query, and search Agent Experience Packs with :memory search specmem query.
See Memory Commands and Agent Memory Layer for provider setup and the full API.
The GEPA skill optimizer is intentionally staged and review-first:
superqode skills optimize --engine geparequires--harness,--tasks, and--live.- SuperQode passes the complete
SKILL.mdtext as the GEPA candidate. - Each candidate is evaluated by creating a temporary harness variant and
running
harness evalagainst the supplied task file. - The evaluator returns per-task score plus side information: prompt, expected checks, task status, failure reason, and failure digest.
- The best candidate is written to
staged/best_skill.md; the live skill is never overwritten by the optimizer. superqode skillopt checkis the bounded-edit gate used before adoption.
The CLI currently exposes the main GEPA search controls: metric-call budget, candidate proposal budget, reflection cost cap, minibatch size, worker count, candidate-selection strategy, frontier type, acceptance criterion, evaluation cache, and merge proposals.
Developer extension points that should stay out of the user-facing workflow until they have a clear review UX:
- GEPA refiner mode
- custom GEPA callbacks from the CLI
- custom reflection prompt templates from the CLI
- multi-component optimization across multiple skills
- automatic adoption of staged skills
Codex SDK:
:codex
:codex status
:codex models
:codex model
:codex effort
:codex sandbox
:codex review
:codex compact
:codex thread
:codex sessions
:codex resume <thread-id>
:codex fork <thread-id>
:codex rename <name>
:codex archive <thread-id>
:codex account
:codex logout
Claude Agent SDK:
:claude
:claude status
:claude model
:claude permission
:claude sessions
:claude resume <session-id>
:claude rename <name>
:claude tag <tag>
:claude commands
:claude command <name> [args]
:claude review
Antigravity CLI handoff:
:antigravity
:antigravity status
:antigravity migrate
:agy status
One-shot coding task:
superqode --print "fix the failing test and summarize the change"Use a vendor/runtime path:
superqode --runtime codex-sdk --print "review this repo"
superqode --runtime claude-agent-sdk --print "summarize the last change"Run with a portable harness spec:
superqode harness run --spec harness.yaml --prompt "implement the smallest safe fix"
superqode harness doctor --spec harness.yaml- One portable harness contract across local, BYOK, ACP, and SDK runtimes.
- Session tree, fork, import, export, and local share artifacts.
- Project trust before enabling project-local plugins/MCP.
- MCP diagnostics and attachable MCP resources.
- Optional Vim-style TUI commands without forcing Vim mode.
- Local-model UX with tool-support detection and provider smoke tests.