@goodandready/dsh-agent-loop-guard 0.2.2

Host-only DeepSeek Harness bundle that prevents tool-call loops without changing DSH core.

Behaviour

• treats a repeated call as a loop only when no successful result or state change has appeared since the previous attempt;
• permits legitimate iterations with the same arguments when the result or explicit progress token changes;
• allows read -> edit -> read -> edit when each step produces new evidence;
• keeps Gitea/curl operations distinct by their complete operation arguments, including HTTP method and endpoint, even when they share a base URL;
• uses maxToolAttemptsPerTurn and maxProgressToolCallsPerTurn as budgets since the last productive action, so productive work resets the counters;
• after a loop or budget denial, enters answer-only mode for the current turn and returns a normal DSH tool denial that requires a text answer; the next turn resets that mode;
• logs every LOOP_GUARD_STOP, LOOP_GUARD_LIMIT, LOOP_GUARD_PROGRESS_LIMIT, LOOP_GUARD_DUPLICATE, and LOOP_GUARD_REPEAT event with a redacted call summary and progress context;
• preserves the assistant-output guard: it detects text-only loops across block/step/turn boundaries and cancels with keepInbox: false; both repeated individual lines and repeated multi-line blocks are covered.

The repeat threshold remains controlled by maxCallsPerRepeatGroup (default 5), but it is evaluated against the current progress epoch rather than raw call count. A successful result is considered productive when its result fingerprint or explicit progress token differs from the last successful evidence. Failed or unknown results do not reset the guard.

The legacy maxCallsPerToolPerTurn setting remains accepted as a compatibility alias. maxToolAttemptsPerTurn: 0 disables only the aggregate no-progress budget; repeat, stop, progress, and assistant-output protections remain active. Denials use the documented tools.guard API and remain normal structured DSH tool results, preserving session persistence.

Verification

npm test npm run check

Assistant output settings

The output guard uses maxRepeatedAssistantLines for single-line repetition and maxRepeatedAssistantBlocks (default 5) for identical multi-line blocks. maxAssistantBlockChars (default 16384) bounds the captured block fingerprint. Streaming chunks and their final assistant/message are deduplicated.