feat(agents): add kimi adapter

[yyovil]

Adds the kimi harness, stacked on #119 (agent platform). Adapter package + Constructors() registration + resolver test.

🤖 Generated with Claude Code

Stack

1. feat(agents): agent platform — registry, activity hooks, harness allowlist #119
2. feat(agents): add grok adapter #120
3. feat(agents): add droid adapter #121
4. feat(agents): add amp adapter #122
5. feat(agents): add agy adapter #123
6. feat(agents): add crush adapter #124
7. feat(agents): add aider adapter #125
8. feat(agents): add cursor adapter #126
9. feat(agents): add qwen adapter #127
10. feat(agents): add copilot adapter #128
11. feat(agents): add goose adapter #129
12. feat(agents): add auggie adapter #130
13. feat(agents): add continue adapter #131
14. feat(agents): add devin adapter #132
15. feat(agents): add cline adapter #133
16. feat(agents): add kimi adapter #134 👈 current
17. feat(agents): add kiro adapter #135
18. feat(agents): add kilocode adapter #136
19. feat(agents): add vibe adapter #137
20. feat(agents): add pi adapter #138
21. feat(agents): add autohand adapter #139

[greptile-apps]

Greptile Summary

This PR adds the Kimi (Moonshot AI) CLI agent adapter as a Tier C harness, following the same structural pattern as the existing grok, cline, and other adapters in the registry.

• kimi/kimi.go: Implements ports.Agent with binary resolution (PATH + common install locations, cross-platform), permission-mode mapping to --auto/-y flags, and intentional no-ops for GetAgentHooks and SessionInfo until Kimi exposes a hook surface.
• kimi/kimi_test.go: Covers manifest, config spec, prompt delivery, all permission modes, restore-command cases, and context cancellation for every method except GetLaunchCommand.
• registry/registry.go + wiring_test.go: Single-line kimi.New() registration and corresponding resolver assertion, consistent with how all other adapters are wired.

Confidence Score: 5/5

Safe to merge; the adapter is well-isolated, no-ops are clearly documented, and the change follows established patterns throughout the codebase.

The new Kimi adapter is a straightforward addition with no impact on existing adapters or critical paths. Binary resolution, permission mapping, and session restore logic are all consistent with peer adapters. The one inconsistency — GetLaunchCommand skipping the early context guard — is a minor behavioral edge case that only surfaces when the binary is already cached on a cancelled context, not a data-loss or security concern.

backend/internal/adapters/agent/kimi/kimi.go — specifically the missing ctx.Err() guard at the top of GetLaunchCommand.

Important Files Changed

Filename	Overview
backend/internal/adapters/agent/kimi/kimi.go	New Tier C adapter for the Kimi CLI; clean structure following peer adapters, but GetLaunchCommand skips the ctx.Err() early-return guard present on every other exported method.
backend/internal/adapters/agent/kimi/kimi_test.go	Thorough unit tests covering permission modes, restore-command, no-op hooks, and context cancellation; TestContextCancellation omits GetLaunchCommand, leaving the cached-binary path untested for cancellation.
backend/internal/adapters/agent/registry/registry.go	Single-line addition of kimi.New() to Constructors(); follows the established registration pattern with no issues.
backend/internal/daemon/wiring_test.go	Adds domain.HarnessKimi/"kimi" to the resolver test table; correct and consistent with how other adapters are verified.

Sequence Diagram

sequenceDiagram
    participant Caller
    participant Plugin as kimi.Plugin
    participant Cache as resolvedBinary cache
    participant Resolve as ResolveKimiBinary

    Caller->>Plugin: GetLaunchCommand(ctx, cfg)
    Plugin->>Cache: resolvedBinary set?
    alt binary cached
        Cache-->>Plugin: return cached path
    else not cached
        Plugin->>Resolve: ResolveKimiBinary(ctx)
        Resolve-->>Plugin: binary path / error
        Plugin->>Cache: store resolved path
    end
    Plugin->>Plugin: appendApprovalFlags(cmd, cfg.Permissions)
    Plugin->>Plugin: append -p prompt (if set)
    Plugin-->>Caller: "[]string{binary, [flags], -p, prompt}"

    Caller->>Plugin: GetRestoreCommand(ctx, cfg)
    Plugin->>Plugin: ctx.Err() check
    Plugin->>Plugin: TrimSpace(agentSessionID)
    alt no agentSessionID
        Plugin-->>Caller: nil, false, nil
    else has agentSessionID
        Plugin->>Cache: resolvedBinary set?
        Plugin->>Plugin: appendApprovalFlags + --session id
        Plugin-->>Caller: "[]string{binary, [flags], --session, id}, true, nil"
    end

Loading

_{Reviews (2): Last reviewed commit: "feat(agents): add kimi adapter" | Re-trigger Greptile}

[greptile-apps]

yyovil has reached the 50-review limit for trial accounts. To continue receiving code reviews, upgrade your plan.

[harshitsinghbhandari]

Verified against upstream Kimi Code docs and shipped binary (v1.37.0), then pushed a fix on top of ce6018e.

Findings re-verified (confirmed):

Two flag-conflict rules from https://moonshotai.github.io/kimi-code/en/reference/kimi-command.html:

"--prompt cannot be used with --yolo, --auto, or --plan — non-interactive mode uses auto permission by default"

"--yolo and --auto cannot be used together with --continue or --session — resumed sessions inherit the approval settings of the original session"

Binary-level check (kimi --help, v1.37.0):

• --yolo / -y (aliases --yes, --auto-approve) — auto-approve all actions
• --prompt / -p (alias --command / -c) — provide prompt non-interactively
• --plan — plan mode
• --auto is not present on this binary at all — kimi --auto … errors with No such option: --auto (Possible options: --agent, --auto-approve, --quiet). So the existing --auto mapping was doubly broken on v1.37.0.

The previous adapter unconditionally prepended --auto or -y before both -p (launch) and --session (restore), violating both rules. Every non-interactive launch and every resume would be rejected at startup.

Changes (backend/internal/adapters/agent/kimi/):

• kimi.go:76-95 — GetLaunchCommand: approval flags emitted only on the interactive path (no prompt). The -p <prompt> path is now bare.
• kimi.go:108-128 — GetRestoreCommand: never emits approval flags; resumed sessions inherit the original session's approval settings. Updated docstring to note cfg.Permissions is intentionally ignored here.
• kimi.go:148-162 — appendApprovalFlags docstring spells out it's interactive-only.
• kimi_test.go — two new table tests assert no --auto/-y/--yolo/--yes/--auto-approve/--plan leaks onto the -p launch or the --session restore for any PermissionMode. The interactive launch mapping is preserved and still verified.

Validation:

• go build ./... — clean
• go test -race ./... — 940 tests / 56 packages, 0 failures

Pushed with --force-with-lease (cleanly fast-forwarded from ce6018e). No design changes, no touch outside kimi/.