slate autoresearch init target

Objective: Add Slate Autoresearch target initialization and document quality-gap/finalization flows; done when generated skill mirrors and target commands verify.

Goal plan: docs/plans/2026-06-01-slate-autoresearch-init-target.md

Completion threshold:

• pnpm slate:ar:setup-target -- <target-id> remains read-only setup-plan.
• pnpm slate:ar:init-target -- <target-id> runs real Autoresearch setup.
• slate-autoresearch covers quality-gap loops and finalization mode.
• Generated .agents/skills/** and .claude/skills/** mirrors are synced.
• Verification commands below pass or have an explicit tool-crash caveat.

Verification surface:

• pnpm install
• node --check tooling/scripts/bench-targets.mjs
• node --check tooling/scripts/slate-autoresearch.mjs
• pnpm bench:targets:check
• pnpm bench:targets:report:check
• pnpm bench:targets:dry-run -- react-active-typing-breakdown
• pnpm slate:ar:setup-target -- react-active-typing-breakdown
• pnpm slate:ar:init-target -- react-active-typing-breakdown
• pnpm slate:ar:prompt-plan -- --prompt "Slate pagination perf target"
• pnpm slate:ar:finalize-preview
• rg audit for generated skill mirrors
• git diff --check

Constraints:

• Edit .agents/rules/slate-autoresearch.mdc, not generated SKILL.md.
• Keep .tmp/slate-v2 as the Slate target workspace.
• Do not commit, push, branch-finalize, or create PRs.
• Keep setup-target safe/read-only; real mutation belongs to init-target.

Boundaries:

• Source of truth: .agents/rules/slate-autoresearch.mdc, tooling/scripts/bench-targets.mjs, tooling/scripts/slate-autoresearch.mjs, package.json, and benchmarks/targets/README.md.
• Allowed edit scope: Slate Autoresearch command/rule/docs support only.
• Browser surface: N/A; this is CLI/agent workflow.
• Tracker sync: N/A; no issue/PR requested.
• Non-goals: no pagination runtime changes, benchmark tuning, or PR branch work.

Blocked condition: Blocked only if the local ../codex-autoresearch CLI is missing or rejects real setup for registered Slate targets.

Start Gates:

Gate	Applies	Evidence
Skill analysis before edits	yes	Read slate-autoresearch and autogoal skills.
Active goal checked or created	yes	Created active goal for init/finalization/quality-gap contract.
Source of truth read before edits	yes	Read rule, package scripts, target helper, target README, and AR CLI help.
Agent-native source/mirror boundary	yes	Edited .agents/rules/slate-autoresearch.mdc, then ran pnpm install.
Browser tool decision	no	N/A: no browser behavior changed.

Work Checklist:

• Added pnpm slate:ar:init-target -- <target-id> for real setup.
• Kept pnpm slate:ar:setup-target -- <target-id> read-only.
• Added pnpm slate:ar:setup-plan-target as the explicit read-only alias.
• Added root Slate AR wrapper that strips pnpm's -- separator and injects .tmp/slate-v2.
• Added quality-gap, research, and finalization package shortcuts.
• Documented quality-gap mode and quality_gap=0 semantics.
• Documented finalization mode, evidence filtering, warnings, and approval boundaries.
• Synced generated skill mirrors with pnpm install.
• Restored active .tmp/slate-v2 Autoresearch config to react-active-typing-breakdown after smoke quality-gap setup.
• Recorded Biome stack overflow as a tooling caveat.

Completion Gates:

Gate	Applies	Required action	Evidence
Named verification threshold	yes	Run command surface and sync checks	Verification evidence below.
Agent rules or skills changed	yes	Run pnpm install and audit mirrors	pnpm install; rg mirror audit passed.
Package scripts changed	yes	Parse package and verify commands	node -e JSON.parse(...); command smoke tests passed.
JS tooling changed	yes	Syntax check scripts	node --check passed for both scripts.
Target registry command contract	yes	Check target registry and dry-run setup	bench:targets:check, report:check, dry-run passed.
Browser surface changed	no	N/A	CLI-only workflow.
Final lint	partial	Run scoped Biome or record crash	Biome 2.3.6 stack-overflowed even on edited JS; syntax and diff checks passed.
Goal plan complete	yes	Run autogoal completion check	check-complete.mjs passed.

Phase / pass table:

Phase	Status	Evidence	Next
Intake and source read	completed	Read relevant skills, scripts, docs, CLI help	implementation
Implementation	completed	Patched rule, scripts, package commands, README	verification
Verification	completed	Commands listed below	closeout
Closeout	completed	Plan filled, goal ready to close	final response

Verification evidence:

• pnpm install synced generated skills.
• node --check tooling/scripts/bench-targets.mjs passed.
• node --check tooling/scripts/slate-autoresearch.mjs passed.
• pnpm bench:targets:check passed with 23 targets.
• pnpm bench:targets:report:check passed.
• pnpm bench:targets:dry-run -- react-active-typing-breakdown passed.
• pnpm slate:ar:setup-target -- react-active-typing-breakdown returned setup-plan JSON and setup-plan is read-only.
• pnpm slate:ar:init-target -- react-active-typing-breakdown created/kept real Autoresearch session files and initialized typing_seconds.
• pnpm slate:ar:state reports react-active-typing-breakdown and typing_seconds.
• pnpm slate:ar:prompt-plan -- --prompt "Slate pagination perf target" proved the wrapper strips pnpm's -- separator.
• pnpm slate:ar:finalize-preview now prints compact output and reports blocked finalization warnings without dumping thousands of files.
• rg audit found quality-gap/finalization/init-target text in source and both generated mirrors.
• Agent-native review found the action surface discoverable: package scripts, source rule, generated skill mirror, and target README all expose the new commands.
• git diff --check passed.
• Caveat: pnpm exec biome check ... --fix crashed with a Biome 2.3.6 stack overflow, including when scoped to the edited JS files.

Reboot status:

Question	Answer
Where am I?	Closeout
Where am I going?	Final response
What is the goal?	Slate Autoresearch init/quality-gap/finalization command contract
What have I learned?	Direct AR shortcuts need a wrapper to strip pnpm's --; raw finalization preview is too noisy for agent use.
What have I done?	Implemented commands, docs, wrapper, generated mirror sync, and command verification.

Open risks:

• Biome stack overflow blocks lint confirmation for the edited JS files. Syntax, JSON parsing, command smoke tests, generated sync, and whitespace checks pass.