| name | isolated-devtools-session |
| description | Canonical bootstrap for isolated Script Kit GPUI DevTools sessions: one front door script (classify, verify-script, start, prove, cleanup), repo-level helpers under scripts/agentic/, bounded timeouts, JSON stdout, progress stderr. Use when agents need parallel DevTools workers, isolated runtime proof, SK_VERIFY script gates, or dev-watch reuse without starting a second GPUI under ./dev.sh. |
| disable-model-invocation | true |
Isolated DevTools Session
Primary entry: scripts/agentic/devtools-session.sh — agents must use this front door instead of ad-hoc session.sh start + unbounded agent-cargo.sh build.
Pair with $script-kit-devtools for inspect/act primitives after bootstrap.
Front Door (required)
bash scripts/agentic/devtools-session.sh classify \
--script kit-init/examples/scripts/todo-app.ts \
--mode auto
bash scripts/agentic/devtools-session.sh verify-script \
--script kit-init/examples/scripts/todo-app.ts
SESSION="dt-agent-$(date +%s)"
bash scripts/agentic/devtools-session.sh start \
--session "$SESSION" \
--mode isolated \
--build never \
--ready-timeout-sec 60 \
--prove \
--cleanup-on-fail
bash scripts/agentic/devtools-session.sh prove --session "$SESSION"
bash scripts/agentic/devtools-session.sh cleanup --session "$SESSION"
Modes
| Mode | When | Cargo | GPUI |
|---|
script-only | SK_VERIFY script proof only | Skip | Skip |
reuse-dev-watch | ./dev.sh running, healthy dev-watch | Skip | Attach only |
isolated | Parallel agent / post-Rust proof | Optional (--build) through target-agent/pools/<pool> | New session from staged binary |
Rules:
- Never start
isolated while ./dev.sh cargo-watch is running (exit 11).
- Never accept
session.sh start with ready:false as done — the front door runs wait-session-ready.sh (60s visible gate; internal start timeout 5s).
- Never run unbounded
agent-cargo.sh build — use build-isolated-binary.sh (120s default) or --build auto|always|never on start.
agent-cargo.sh defaults to the bounded target-agent/pools/agent-debug pool with a visible lock; use SCRIPT_KIT_AGENT_TARGET_MODE=exclusive only when a task truly needs its own cache.
- Isolated builds stage
target-agent/runtime/<session>/script-kit-gpui and sessions launch that staged binary via SCRIPT_KIT_GPUI_BINARY; agents must not promote into target/debug/script-kit-gpui.
Reuse dev-watch
bash scripts/agentic/devtools-session.sh start \
--mode reuse-dev-watch \
--session dev-watch \
--build never
Thin Skill Wrappers
Scripts under .agents/skills/isolated-devtools-session/scripts/ are exec wrappers to scripts/agentic/:
preflight-isolated.sh
wait-session-ready.sh
build-isolated-binary.sh
start-isolated.sh
Do not duplicate logic in the skill folder.
Verification
bash scripts/agentic/verify-devtools-session.sh
./scripts/agentic/agent-cargo.sh test devtools_session_contract --
Environment
| Variable | Purpose |
|---|
SCRIPT_KIT_SESSION_DIR | FIFO/state root (default /tmp/sk-agentic-sessions) |
SCRIPT_KIT_AGENT_ID | Agent identity for logs and explicit exclusive targets |
SCRIPT_KIT_CARGO_TARGET_POOL | Named shared agent target pool (default agent-debug) |
SCRIPT_KIT_AGENT_TARGET_MODE | pool by default; set exclusive only for per-agent caches |
SCRIPT_KIT_GPUI_BINARY | Staged binary path for isolated session launch |
SCRIPT_KIT_TEST_NOTES_DB_PATH | Sandbox notes DB (--notes-sandbox on start) |
Anti-Patterns
./dev.sh + isolated session.sh start (second GPUI → empty app.log, RPC hangs).
- Bare
cargo build while ./dev.sh holds target/.cargo-lock.
- Full GPUI bootstrap before
verify-script for Bun-only demos.
SCRIPT_KIT_SESSION_READY_TIMEOUT_MS=60000 inside start and a 60s wait loop (double silence) — use the front door defaults.
Pairing
$script-kit-devtools — investigation after bootstrap.
$protocol-automation — RPC semantics.
$agentic-testing — regression recipes after isolation.