oikos/cmd/nomos/continue.go

package main

import (
	"context"
	"encoding/json"
	"fmt"
	"log/slog"
	"regexp"
	"strings"
	"time"

	"github.com/dtoro/oikos/internal/safego"
	"github.com/google/uuid"
)

// execIDRe matches "execution <uuid>" in a tool result — the phrasing shared
// by request_execution / run when they queue or start a gated execution.
// Only these async executions need continuation; the synchronous auto-run
// path returns its output inline and is already observed in-turn.
var execIDRe = regexp.MustCompile(`(?i)execution\s+([0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12})`)

func extractExecutionIDs(toolResult string) []uuid.UUID {
	matches := execIDRe.FindAllStringSubmatch(toolResult, -1)
	seen := map[uuid.UUID]bool{}
	var out []uuid.UUID
	for _, m := range matches {
		if id, err := uuid.Parse(m[1]); err == nil && !seen[id] {
			seen[id] = true
			out = append(out, id)
		}
	}
	return out
}

// runContinuationWorker is the event loop that replaces the human typing
// "continue". It polls for gated executions that (a) were initiated by a chat
// session and (b) have just finished, and — while that agent has an open assent
// window (an approved plan is in flight) — feeds each result back into the
// agent so it proceeds to the next step or recovers from the failure, all
// without an operator tick. Blocks until ctx is cancelled.
func (a *agent) runContinuationWorker(ctx context.Context) {
	if a.store == nil {
		slog.Warn("nomos: continuation worker disabled (no store)")
		return
	}
	slog.Info("nomos: continuation worker started")
	ticker := time.NewTicker(4 * time.Second)
	defer ticker.Stop()
	for {
		select {
		case <-ctx.Done():
			return
		case <-ticker.C:
			a.processContinuations(ctx)
		}
	}
}

// processContinuations dispatches each pending item as its OWN goroutine
// (safego.Go, so a panic deep in one task's resumed turn — JSON parsing of
// model output, an unexpected nil in a tool result — is recovered and logged
// instead of taking down this whole function, which used to run every
// item sequentially in the SAME goroutine as the ticker loop. Two problems
// that fixed: (1) throughput — task B's continuation no longer waits for
// task A's full (up to 10-minute) resumed turn to finish first, the exact
// per-task blocking this session's earlier concurrency work removed from the
// live-chat path but had left in place here; (2) survivability — since Go
// panics unwind the goroutine they occur in, an unrecovered one here used to
// mean this call (and every future tick, since the whole ticker loop runs in
// one goroutine) would simply stop — auto-continuation for every task would
// silently die until nomos restarted. Now a single bad item can only ever
// take down its own goroutine.
func (a *agent) processContinuations(ctx context.Context) {
	pending := a.store.pendingContinuations(ctx, 5)
	for _, p := range pending {
		// Scope gate: only auto-continue while an approved plan is active FOR
		// THIS SESSION. Checked per-item, not once for the whole batch — with
		// multiple tasks in flight, one task's open window must never cover a
		// pending continuation belonging to a different task.
		if !a.store.assentWindowActive(ctx, a.agentID, p.SessionID) {
			// A finished one-off execution with no window is left as-is
			// (marked continued so we don't re-check it forever) — the
			// operator decides what happens next, as today.
			a.store.markContinued(ctx, p.ExecID)
			continue
		}
		a.store.markContinued(ctx, p.ExecID) // stamp first: a failure here must not cause a re-continue loop
		safego.Go("nomos:continue-session:"+p.SessionID, func() { a.continueSession(ctx, p) })
	}
}

// continueSession re-invokes the agent for one finished execution. Persists
// progress LIVE — a placeholder row immediately, updated in place as each
// tool call completes — instead of only saving once the whole continuation
// finishes. The frontend polls (see chat.ts startPolling); without
// incremental persistence here, a continuation that runs several tool calls
// before concluding would look like total silence in the UI for however long
// that takes, which is exactly the "I just wait while nothing happens"
// complaint this exists to fix — polling alone only helps if there's
// something new to poll for.
func (a *agent) continueSession(ctx context.Context, p pendingContinuation) {
	slog.Info("nomos: auto-continuing session", "session", p.SessionID, "execution", p.ExecID, "status", p.Status)
	a.resumeSession(ctx, p.SessionID, buildContinuationNote(p))
}

// resumeSession re-invokes the agent for a session with a system-injected note —
// a finished execution (continueSession) or an operator's answer to a question
// (handleAnswerQuestion) — persisting progress LIVE (a placeholder row updated
// in place as each tool call lands) so the frontend poller sees each step,
// instead of total silence until the whole resume concludes.
func (a *agent) resumeSession(ctx context.Context, sessionID, note string) {
	placeholder, _ := json.Marshal(map[string]any{
		"role": "assistant",
		"text": "",
		"auto": true,
	})
	msgID, err := a.store.insertMessageReturningID(ctx, sessionID, "assistant", placeholder)
	if err != nil {
		slog.Error("nomos: resume placeholder insert failed", "session", sessionID, "error", err)
	}

	var toolCalls []map[string]any
	var finalText, errText string

	persist := func() {
		if msgID == uuid.Nil {
			return
		}
		text := finalText
		if text == "" && errText != "" {
			text = fmt.Sprintf("(auto-continuation hit an internal error and did not respond: %s — the execution's own result is above; you may need to prompt the agent again)", errText)
		}
		body, _ := json.Marshal(map[string]any{
			"role":       "assistant",
			"text":       text,
			"tool_calls": toolCalls,
			"auto":       true, // marks this as an autonomous continuation, not an operator turn
		})
		a.store.updateMessage(ctx, msgID, body)
	}

	// One retry if the LLM call itself produced nothing (transient flake /
	// empty-response) — the whole point of this mechanism is "don't give up
	// on the first error," which should apply to the continuation call
	// itself, not just the homelab commands it's continuing. Found live: a
	// destructive-recovery continuation hit an empty LLM response, its
	// internal retry (chatWith's own maxLLMRetries=1) also came up empty, and
	// without this outer retry the operator would see nothing at all.
	cctx, cancel := context.WithTimeout(ctx, 10*time.Minute)
	defer cancel()
	for attempt := 0; attempt < 2; attempt++ {
		toolCalls, finalText, errText = nil, "", ""
		emit := func(ev agentEvent) {
			if ev.Type == "tool_use" || ev.Type == "tool_result" {
				if m, ok := ev.Data.(map[string]any); ok {
					m["type"] = ev.Type
					toolCalls = append(toolCalls, m)
				}
				persist() // live: a poller sees this step land within seconds
			}
			if ev.Type == "text" {
				finalText, _ = ev.Data.(string)
			}
			if ev.Type == "error" {
				errText, _ = ev.Data.(string)
			}
		}
		a.chatWith(cctx, sessionID, "", note, emit)
		if finalText != "" || len(toolCalls) > 0 {
			break
		}
		if attempt == 0 {
			slog.Warn("nomos: resume produced nothing, retrying once", "session", sessionID, "error", errText)
		}
	}

	if errText != "" && finalText == "" {
		slog.Error("nomos: resume produced no response after retry", "session", sessionID, "error", errText)
		// Give the task a real, operator-visible terminal state instead of
		// leaving it silently stuck at whatever status it was in (typically
		// 'executing' or 'awaiting_input') forever. Before this, a
		// permanently-failed resume was invisible beyond a log line — the
		// task board just showed a task that never changed, with nothing
		// telling the operator it needed attention. Marking it failed here
		// doesn't prevent the operator from continuing to work the task via
		// a fresh chat message afterward; it just stops the silent hang.
		summary := fmt.Sprintf("Auto-resume failed after retrying: %s", errText)
		if len(summary) > 200 {
			summary = summary[:200] + "…"
		}
		if cerr := a.store.completeTask(context.Background(), sessionID, "failure", summary); cerr != nil {
			slog.Error("nomos: failed to mark task failed after resume gave up", "session", sessionID, "error", cerr)
		}
	}
	persist() // final state — same row, updated one last time with the concluding text
}

// buildContinuationNote frames the finished execution for the model: what
// happened, and what to do about it. The persist-through-errors instruction
// lives here (and in SOUL) so the agent recovers instead of stopping.
func buildContinuationNote(p pendingContinuation) string {
	action := p.Action
	if i := strings.IndexByte(action, ':'); i > 0 && len(action) > 40 {
		action = action[:i] // keep just the action verb for brevity; params are in the DB
	}
	result := p.Result
	if len(result) > 3000 {
		result = result[:3000] + "…[truncated]"
	}
	var b strings.Builder
	fmt.Fprintf(&b, "[System: execution %s (%s) finished with status=%s.\nResult: %s\n\n",
		p.ExecID, action, p.Status, result)
	switch p.Status {
	case "completed":
		b.WriteString("It SUCCEEDED. Continue the approved plan: run the next step. If this was the final step, verify the end goal actually works (e.g. curl the service) and then report success to the operator. Do NOT stop and wait for the operator to say 'continue'.")
	case "failed", "cancelled":
		b.WriteString("It FAILED. Do NOT give up or hand back to the operator. Diagnose the cause from the result above (and by running read-only inspection commands if needed), form a hypothesis, fix it, and retry or take an alternative approach. You have an active assent window, so config_mutation steps run without re-approval. Only stop and ask the operator if you are genuinely blocked (need information only they have) or the fix would require a destructive action they haven't approved.")
	default: // denied / revoked
		b.WriteString("The operator denied or revoked this step. Stop executing this plan and briefly acknowledge.")
	}
	b.WriteString("]")
	return b.String()
}