You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Close the user-facing feature gap between go-code and MoonshotAI's kimi-code CLI/TUI, as identified in the 2026-07-19 comparison review. This is the tracking and sequencing issue for the whole program: 19 child epics in 3 tiers, each containing PR-sized slices.
Areas where go-code already leads (do not regress these, and do not build parity items for them): rollout replay/fork/drift + forensics, container/VM/worktree workspace backends, GitHub/Slack/Linear webhooks, scoped API keys + multi-tenancy, capability profiles, the deferred find_tool system, goals with dependencies, and the trainer/improve loop.
Workflow (how this program executes)
Pickup order is tier order, then in-tier order. Tier 1 = high-impact UX where server pieces mostly exist; Tier 2 = strategic differentiators; Tier 3 = cheap polish.
One slice = one PR. Slices inside each epic are ordered by dependency; slices marked "parallel-safe" may be picked concurrently.
Every slice follows the repo runbooks: bootstrap a worktree with scripts/init.sh <slug>, strict TDD per docs/runbooks/testing.md (no red baseline tolerated), update affected docs and folder indexes, then merge with scripts/verify-and-merge.sh.
Cross-epic dependencies are listed in each epic's ## Dependencies section — do not start an epic before its dependencies merge.
All 19 child epics created with PR-sized slices and the repo template format
All 19 child epics closed (slices merged to main)
Test Requirements
Per-slice, per the runbooks: behavior tests written first, GOCACHE=/tmp/go-build ./scripts/test-regression.sh green before merge. This tracking issue itself requires no tests.
Documentation Updates Required
Each child epic lists its own doc updates. When the program completes, update README.md ("What You Get") and website/docs/cli/tui.md so the public feature list reflects the new capabilities.
Dependencies
None — this is the root tracker. Inter-epic dependencies are recorded inside each child epic.
Objective
Close the user-facing feature gap between
go-codeand MoonshotAI's kimi-code CLI/TUI, as identified in the 2026-07-19 comparison review. This is the tracking and sequencing issue for the whole program: 19 child epics in 3 tiers, each containing PR-sized slices.Context
Comparison basis: kimi-code official docs (slash commands, keyboard, tools,
kimicommand, interaction, sessions, goals, hooks, plugins, themes, agents, MCP, skills, changelog) versus an inventory ofcmd/harnesscli,cmd/harnessd, andinternal/onmain.Areas where go-code already leads (do not regress these, and do not build parity items for them): rollout replay/fork/drift + forensics, container/VM/worktree workspace backends, GitHub/Slack/Linear webhooks, scoped API keys + multi-tenancy, capability profiles, the deferred
find_toolsystem, goals with dependencies, and the trainer/improveloop.Workflow (how this program executes)
scripts/init.sh <slug>, strict TDD perdocs/runbooks/testing.md(no red baseline tolerated), update affected docs and folder indexes, then merge withscripts/verify-and-merge.sh.## Dependenciessection — do not start an epic before its dependencies merge..github/ISSUE_TEMPLATE/task.md) extended withExisting Foundation(verified file citations) andProposed PR Slicessections, matching the house epic style of epic(hooks): config-driven lifecycle hooks (shell/HTTP) adapted onto existing harness hook interfaces #737.Child Epics
Tier 1 — high-impact UX, server pieces mostly exist
epic(tui): mid-turn steering from the TUIepic(tui): shell mode (! prefix) with background handoff and historyepic(tui): image input — clipboard paste and ReadMediaFile toolepic(harness): /undo — remove recent prompts from the active contextepic(harness): /fork — fork the live session with full historyepic(tui): unified /tasks background panelepic(harness): complete plan mode — plan file, restricted writes, exit approval with optionsTier 2 — strategic differentiators
epic(server): read-only session visualizer served by harnessdepic(acp): Agent Client Protocol server mode for Zed/JetBrainsepic(subagents): swarm fan-out tool with template, resume, and live progress panelepic(plugins): plugin manifest, GitHub install, trust, and /plugins managerTier 3 — polish
epic(tui): theme system — token schema, JSON theme files, picker, hot reloadepic(tui): /compact [instruction] and a visible compaction summaryepic(config): /reload — apply config changes without restartepic(tui): quality-of-life commands — /title, /init, /add-dir, /feedback, upgradeepic(skills): skill argument placeholders and slash-command namespacingepic(mcp): OAuth login flow for remote MCP serversepic(cli): headless exit-code conventions for runs and goalsepic(install): OS-service install for harnessd (launchd/systemd)Acceptance Criteria
main)Test Requirements
Per-slice, per the runbooks: behavior tests written first,
GOCACHE=/tmp/go-build ./scripts/test-regression.shgreen before merge. This tracking issue itself requires no tests.Documentation Updates Required
Each child epic lists its own doc updates. When the program completes, update
README.md("What You Get") andwebsite/docs/cli/tui.mdso the public feature list reflects the new capabilities.Dependencies
None — this is the root tracker. Inter-epic dependencies are recorded inside each child epic.