Evidence snapshot

CLI Host Compatibility

Version-pinned evidence for six agent CLIs hosted by Kota over PTY.

Not a ranking. Unknown means unverified, not unsupported.

ObservedAugust 2, 2026 · PDT

PlatformmacOS 26.5.2 arm64

Kotav0.1.8 · 156835c

Providers6 hosted CLIs

Records66 evidence cells

Comparison matrix

What Kota has verified

VERIFIED 25PARTIAL 33UNKNOWN 5NOT TESTED 0NOT EXPOSED 3

Capability	Claude Code `claude` · 2.1.220	Codex `codex` · 0.146.0	Antigravity CLI `agy` · 1.1.9	OpenCode `opencode` · 1.17.9	Pi `pi` · 0.80.10	Kimi Code `kimi` · 0.31.0
01 Workspace selection How is the worktree selected?	VERIFIED Process cwd = incarnation worktree. Evidence CODE PATHUNIT FIXTURE PTY launch and argument construction ↗Launch argument fixtures ↗	VERIFIED Process cwd = incarnation worktree. Evidence CODE PATHUNIT FIXTURE PTY launch and argument construction ↗Launch argument fixtures ↗	VERIFIED Process cwd, plus `--add-dir <cwd>`. Evidence CODE PATHUNIT FIXTURE PTY launch and argument construction ↗Launch argument fixtures ↗	VERIFIED `--dir <cwd>`, or a positional project in mini mode. Evidence CODE PATHUNIT FIXTURE PTY launch and argument construction ↗Launch argument fixtures ↗	VERIFIED Process cwd = incarnation worktree. Evidence CODE PATHUNIT FIXTURE PTY launch and argument construction ↗Launch argument fixtures ↗	VERIFIED Process cwd = incarnation worktree. Evidence CODE PATHUNIT FIXTURE PTY launch and argument construction ↗Launch argument fixtures ↗
02 Workspace instruction files Which project files are read?	VERIFIED Kota writes `CLAUDE.md`. `AGENTS.md` is not read automatically. Evidence CODE PATH Workspace adapter selection ↗Claude Code memory ↗	VERIFIED Kota writes `AGENTS.md`; Codex layers files from repo root to cwd. Evidence CODE PATH Workspace adapter selection ↗Codex AGENTS.md ↗	VERIFIED Kota writes `AGENTS.md`; the CLI also reads `GEMINI.md`. Evidence CODE PATH Conflict resolution was not tested separately. Workspace adapter selection ↗Antigravity context files ↗	VERIFIED Kota writes `AGENTS.md`; it wins over same-scope `CLAUDE.md`. Evidence CODE PATH Workspace adapter selection ↗OpenCode rules ↗	VERIFIED Kota writes `AGENTS.md`; Pi gathers it or `CLAUDE.md` toward cwd. Evidence CODE PATH Same-directory tie-break is undocumented. Workspace adapter selection ↗Pi context files ↗	VERIFIED Kota writes `AGENTS.md`; Kimi also reads `.kimi-code/AGENTS.md`. Evidence CODE PATH Workspace adapter selection ↗Kimi Code instruction files ↗
03 Approval / unrestricted launch Which launch flag changes approval?	PARTIAL Default: `--dangerously-skip-permissions`. Evidence CODE PATHUNIT FIXTURE Flag emission is verified; full bypass scope is not. PTY launch and argument construction ↗Launch argument fixtures ↗	PARTIAL Default: `--dangerously-bypass-approvals-and-sandbox`. Evidence CODE PATHUNIT FIXTURE Flag emission is verified; full bypass scope is not. PTY launch and argument construction ↗Launch argument fixtures ↗	PARTIAL Default: `--dangerously-skip-permissions`. Evidence CODE PATHUNIT FIXTURE Flag emission is verified; full bypass scope is not. PTY launch and argument construction ↗Launch argument fixtures ↗	PARTIAL Conditional: `--dangerously-skip-permissions`. Evidence CODE PATHUNIT FIXTURE Mini mode drops unsupported flags; scope stays version-dependent. PTY launch and argument construction ↗Launch argument fixtures ↗	PARTIAL Default: `--approve`. Evidence CODE PATHUNIT FIXTURE Flag emission is verified; full bypass scope is not. PTY launch and argument construction ↗Launch argument fixtures ↗	PARTIAL Default: `--yolo`. Evidence CODE PATHUNIT FIXTURE Flag emission is verified; full bypass scope is not. PTY launch and argument construction ↗Launch argument fixtures ↗
04 Session resume How is a native session resumed?	PARTIAL `--resume <session-id>`. Evidence CODE PATHUNIT FIXTURE Launch shape verified; cross-version resume is CLI-owned. PTY launch and argument construction ↗Launch argument fixtures ↗	PARTIAL `resume <session-id>`. Evidence CODE PATHUNIT FIXTURE Launch shape verified; cross-version resume is CLI-owned. PTY launch and argument construction ↗Launch argument fixtures ↗	PARTIAL `--conversation <conversation-id>`. Evidence CODE PATHUNIT FIXTURE The backend revalidates the conversation. PTY launch and argument construction ↗Launch argument fixtures ↗	PARTIAL `--session <session-id>`, when supported. Evidence CODE PATHUNIT FIXTURE Mini-mode support depends on installed help. PTY launch and argument construction ↗Launch argument fixtures ↗	PARTIAL `--session-id <session-id>`. Evidence CODE PATHUNIT FIXTURE Launch shape verified; cross-version resume is CLI-owned. PTY launch and argument construction ↗Launch argument fixtures ↗	PARTIAL `--session <session-id>`. Evidence CODE PATHUNIT FIXTURE Launch shape verified; cross-version resume is CLI-owned. PTY launch and argument construction ↗Launch argument fixtures ↗
05 Session lifetime / retention Who owns retention, and for how long?	VERIFIED Default: 30 days; configurable with `cleanupPeriodDays`. Evidence CODE PATHLIVE OBSERVATION Kota observed expiry and offers Start Fresh on exact transcript NotFound. Claude expired-session handling ↗Claude Code sessions ↗	UNKNOWN No published TTL; no automatic expiry observed. Evidence History limits do not establish resume TTL.	UNKNOWN No published backend TTL; resume revalidates the conversation. Evidence Local cache does not imply permanent retention.	UNKNOWN No published TTL; no automatic expiry observed.	UNKNOWN No published TTL; no automatic expiry observed.	UNKNOWN No published TTL; no automatic expiry observed.
06 Model selection at launch How is the model chosen?	PARTIAL Non-default: `--model <id>`. Default defers to the CLI. Evidence CODE PATHUNIT FIXTURE Not every listed model is end-to-end verified. Launch argument fixtures ↗	PARTIAL Non-default: `--model <id>`. Default defers to the CLI. Evidence CODE PATHUNIT FIXTURE Not every listed model is end-to-end verified. Launch argument fixtures ↗	PARTIAL Non-default: `--model <id>`. Default defers to the CLI. Evidence CODE PATHUNIT FIXTURE Not every listed model is end-to-end verified. Launch argument fixtures ↗	PARTIAL Non-default: `--model <id>` in the detected launch shape. Evidence CODE PATHUNIT FIXTURE Not every listed model is end-to-end verified. Launch argument fixtures ↗	PARTIAL Non-default: `--model <id>`. Default defers to the CLI. Evidence CODE PATHUNIT FIXTURE Not every listed model is end-to-end verified. Launch argument fixtures ↗	PARTIAL Non-default: `--model <id>`. Default defers to the CLI. Evidence CODE PATHUNIT FIXTURE Not every listed model is end-to-end verified. Launch argument fixtures ↗
07 Reasoning / effort level Can Kota set effort at launch?	PARTIAL Launch flag: `--effort`. Evidence CODE PATHUNIT FIXTURE Values vary by version and model. Launch argument fixtures ↗	PARTIAL Launch config: `model_reasoning_effort=<value>`. Evidence CODE PATHUNIT FIXTURE Values vary by version and model. Launch argument fixtures ↗	NOT EXPOSED Not exposed by Kota. Evidence CODE PATH Launch argument fixtures ↗	NOT EXPOSED Not exposed by Kota. Evidence CODE PATH Launch argument fixtures ↗	PARTIAL Launch flag: `--thinking`. Evidence CODE PATHUNIT FIXTURE Values vary by version and model. Launch argument fixtures ↗	NOT EXPOSED Not exposed by Kota. Evidence CODE PATH Launch argument fixtures ↗
08 Local chat / session records Which native records does Kota read?	VERIFIED `~/.claude/projects/.../*.jsonl` · JSONL. Evidence CODE PATHUNIT FIXTURE Native session source locators ↗	VERIFIED `~/.codex/sessions/*/.jsonl` · JSONL. Evidence CODE PATHUNIT FIXTURE Native session source locators ↗	VERIFIED `~/.gemini/antigravity-cli/brain/<conversation>/.../transcript[_full].jsonl` · JSONL. Evidence CODE PATHUNIT FIXTURE Native session source locators ↗	VERIFIED `~/.local/share/opencode/opencode.db` · SQLite + WAL/SHM. Evidence CODE PATHUNIT FIXTURE Native session source locators ↗Native-source watch plan ↗	VERIFIED `~/.pi/agent/sessions/.../*.jsonl` · JSONL. Evidence CODE PATHUNIT FIXTURE Native session source locators ↗	VERIFIED `${KIMI_CODE_HOME:-~/.kimi-code}` · JSON + main-agent JSONL. Evidence CODE PATHUNIT FIXTURE Kota reads `agents/main/wire.jsonl` only. Native session source locators ↗
09 Record refresh / flush cadence When does new state become visible?	PARTIAL 500 ms Kota watch throttle; CLI flush timing is not guaranteed. Evidence CODE PATH Native-source watch plan ↗	PARTIAL 500 ms Kota watch throttle; turn IDs improve correlation. Evidence CODE PATH Native-source watch plan ↗Native work-state mapping ↗	PARTIAL 500 ms Kota watch throttle; backend flush timing is not guaranteed. Evidence CODE PATH Native-source watch plan ↗	PARTIAL 2 s SQLite monitor; CLI commit timing is not guaranteed. Evidence CODE PATH Native-source watch plan ↗	PARTIAL 500 ms Kota watch throttle; CLI flush timing is not guaranteed. Evidence CODE PATH Native-source watch plan ↗	PARTIAL 500 ms Kota watch throttle; step events improve correlation. Evidence CODE PATHUNIT FIXTURE Native-source watch plan ↗Kimi step lifecycle parser ↗
10 Working-state detection How does Kota infer working state?	PARTIAL Process, delivery, native activity, and a 5-minute fallback. Evidence CODE PATHUNIT FIXTURE Alive, working, and visible remain separate states. Native work-state mapping ↗Agent runtime reconciliation ↗Working-lifecycle tests ↗	PARTIAL Turn IDs, reconciled with process and delivery state. Evidence CODE PATHUNIT FIXTURE No signal is universal readiness. Native work-state mapping ↗Agent runtime reconciliation ↗Working-lifecycle tests ↗	PARTIAL Process, delivery, transcript activity, and a 5-minute fallback. Evidence CODE PATHUNIT FIXTURE Backend state is not a readiness API. Native work-state mapping ↗Agent runtime reconciliation ↗Working-lifecycle tests ↗	PARTIAL Process, delivery, SQLite activity, and a 5-minute fallback. Evidence CODE PATHUNIT FIXTURE Database visibility may lag the PTY. Native work-state mapping ↗Agent runtime reconciliation ↗Working-lifecycle tests ↗	PARTIAL Process, delivery, native activity, and a 5-minute fallback. Evidence CODE PATHUNIT FIXTURE No signal is universal readiness. Native work-state mapping ↗Agent runtime reconciliation ↗Working-lifecycle tests ↗	PARTIAL `step.begin` / `step.end` + turn ID, with runtime fallback. Evidence CODE PATHUNIT FIXTURE Kimi step lifecycle parser ↗Agent runtime reconciliation ↗Working-lifecycle tests ↗
11 Host transport PTY or structured transport?	VERIFIED Interactive child process on a real PTY; no SDK transport. Evidence CODE PATHUNIT FIXTURE PTY launch and argument construction ↗Launch argument fixtures ↗	VERIFIED Interactive child process on a real PTY; no SDK transport. Evidence CODE PATHUNIT FIXTURE PTY launch and argument construction ↗Launch argument fixtures ↗	VERIFIED Interactive child process on a real PTY; no SDK transport. Evidence CODE PATHUNIT FIXTURE PTY launch and argument construction ↗Launch argument fixtures ↗	VERIFIED Real PTY; launch adapts between interactive and mini modes. Evidence CODE PATHUNIT FIXTURE PTY launch and argument construction ↗Launch argument fixtures ↗	VERIFIED Interactive child process on a real PTY; no SDK transport. Evidence CODE PATHUNIT FIXTURE PTY launch and argument construction ↗Launch argument fixtures ↗	VERIFIED Real PTY; CSI-u Enter submits prompts. Evidence CODE PATHUNIT FIXTURE PTY launch and argument construction ↗Kimi prompt submit sequence ↗

Mapping notes

Provider notes

Antigravity CLI: Legacy gemini and gemini-cli config names resolve to Antigravity, not Gemini CLI.
OpenCode: Launch adapts between run --interactive and --mini from installed CLI help.
Kimi Code: Provider alias: kimi-code.

CLI versions describe this snapshot, not every historical fixture. Source data →