Greptile code review

This repo uses Greptile for automated review. Before merge, aim for Confidence Score: 5/5 with zero unresolved review threads — see CONTRIBUTING.md.

Run a review — add a PR comment with:

@greptile review

Give it ~5-10 minutes (sometimes longer) for results, then fix feedback and re-trigger until you reach Confidence Score: 5/5.

Optional: automate with the greploop skill.

Greptile Summary

This PR removes the /list omnibus command and replaces it with per-domain subcommands (/integrations list, /mcp list, /model show, /tools list). It also adds a graceful fallback in the LLM action planner so that context-overflow errors trigger the deterministic mapper instead of propagating as unhandled exceptions.

• /list removal: All /list handler code, interactive menus, and catalog entries are deleted from integrations.py. A new tools_cmds.py wires /tools list using the new make_list_root_handler factory in types.py, which centralises unknown-subcommand error handling and session marking.
• Prompt-overflow fallback (planner.py): PlannerLLMError on context overflow is now caught; the deterministic action mapper is tried first and falls back to assistant_handoff only when it finds nothing actionable. A dedicated test file (test_llm_action_planner_fallback.py) covers Anthropic/OpenAI overflow messages, informational-question preservation, empty-mapper handoff, and non-overflow re-raise.
• LLM payload reduction (slash_catalog.py, prompting.py): The slash catalog is no longer appended to the system prompt; slash_invoke_tool_description now emits only command names to keep prompt size small under strict token budgets.

Confidence Score: 5/5

Safe to merge — the refactoring is mechanical and well-tested, and the new overflow fallback path is covered by a dedicated test file.

All /list callsites and cross-references (deterministic mapper, registry rules, contract YAMLs, test fixtures) are consistently updated to the new /<domain> list form. The make_list_root_handler factory handles both the happy path and the unknown-subcommand error case correctly. The overflow fallback adds a meaningful resilience improvement and its test suite covers the main edge cases, including informational-question preservation that was a concern in a prior review round.

No files require special attention beyond the minor PEP 8 blank-line nit in failure_explain.py.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["User types /tools, /integrations, /mcp, /model"] --> B["dispatch_slash"]
    B --> C{"Command match"}
    C -->|"/tools"| D["make_list_root_handler\n(types.py)"]
    C -->|"/integrations"| E["_cmd_integrations\n(integrations.py)"]
    C -->|"/mcp"| F["_cmd_mcp\n(integrations.py)"]
    D --> G{"sub in aliases?"}
    G -->|"list / ls / tool / tools"| H["_list_tools → render_tools_table"]
    G -->|"unknown"| I["print error\nsession.mark_latest(ok=False)"]

    J["User natural language"] --> K["plan_actions_with_llm_result\n(planner.py)"]
    K --> L["_call_llm"]
    L -->|"PlannerLLMError"| M{"is_context_length_overflow?"}
    M -->|"Yes"| N["_plan_prompt_overflow_fallback"]
    M -->|"No"| O["re-raise"]
    N --> P["map_actions_result\n(deterministic mapper)"]
    P -->|"actions found"| Q["finalize_planner_result_with_trace"]
    P -->|"no actions"| R["_fallback_handoff\n(assistant_handoff)"]
    Q --> S["LlmActionPlanResult\npolicy_trace: fallback_prompt_too_long"]
    R --> S
    L -->|"success"| T["_parse_tool_plan → finalize"]

_{Reviews (7): Last reviewed commit: "refactor(planner): streamline context le..." | Re-trigger Greptile}

@greptile-apps review again

in /agents imo not required, to add agents list its a single command, list will be necessary which has subcommands?

here also not required, could you cross check?

not required

not required

@muddlebee Thanks for the review — agreed these list aliases were unnecessary. I’ve removed /agents list, /sessions list, /tasks list, and /watches list, so these commands now use the direct form only (/agents, /sessions, /tasks, /watches). I also updated the related tests accordingly.

@greptile-apps review again

great work @Devesh36

can you look in to this: the "prompt too long" fallback never runs. In planner.py the guard checks for the exact phrase "prompt too long", but providers don't say that: Anthropic returns prompt is too long: … and OpenAI says …maximum context length is…. Neither matches, so the error just re-raises and the fallback is dead. Suggest matching on "too long" or a small regex. There's also no test for this path.

Dead code: make_list_root_handler (types.py) has no callers : /tools even copies the same logic inline. Wire it up or drop it.

Stale comments: still reference /list tools / /list mcp / /list integrations in tool_catalog.py:13 and rendering.py:59-60.

Description mismatch: the PR says /agents list/ls and /sessions list were added, but they aren't in the code — /agents list still errors and _cmd_sessions ignores its args.

great work @Devesh36

can you look in to this: the "prompt too long" fallback never runs. In planner.py the guard checks for the exact phrase "prompt too long", but providers don't say that: Anthropic returns prompt is too long: … and OpenAI says …maximum context length is…. Neither matches, so the error just re-raises and the fallback is dead. Suggest matching on "too long" or a small regex. There's also no test for this path.

Dead code: make_list_root_handler (types.py) has no callers : /tools even copies the same logic inline. Wire it up or drop it.

Stale comments: still reference /list tools / /list mcp / /list integrations in tool_catalog.py:13 and rendering.py:59-60.

Description mismatch: the PR says /agents list/ls and /sessions list were added, but they aren't in the code — /agents list still errors and _cmd_sessions ignores its args.

nice catch , i'll resolve this asap

@greptile-apps review again

@greptile-apps review again

@greptile-apps review again

@greptile-apps review again

looks good to me

⚡ LGTM → Merged. @Devesh36, your work is in. Every commit counts — thank you for this one.

👋 Join us on Discord - OpenSRE : hang out, contribute, or hunt for features and issues. Everyone's welcome.