nervtimer（nervtimer 效率）

NervTimer

Use this skill when the user wants OpenClaw to set a reminder timer and keep reminding until explicit completion.

This skill is designed to be published via ClawHub and installed with:

openclaw skills install nervtimer

Required behavior

1. Parse user intent into structured timer data.
2. Support one_shot and recurrent schedules.
3. Extract optional reason/purpose text from user intent.
4. When due, start nagging every 5 minutes.
5. Stop nagging only when the user explicitly says done.
6. For recurring timers: done ends only the current nagging phase, not the whole recurring schedule.

Deterministic + generative split

• Deterministic:

- state transitions (scheduled, nagging, done) - nag interval (5 minutes) - escalation stage from nag_count - schedule payload construction

• Generative (LLM):

- final reminder sentence style - personality adaptation - increasing urgency/annoyance within policy bounds

Files in this skill

• {baseDir}/references/intent-schema.md
• {baseDir}/references/escalation-policy.md
• {baseDir}/scripts/validate-intent.sh
• {baseDir}/scripts/state.sh
• {baseDir}/scripts/build-cron-payload.sh

Workflow

Execution rule:

• Always execute skill scripts via bash <script-path> ....
• Do not execute scripts directly via shebang path.
• If a script call fails, immediately fall back to direct cron tool calls and inform the user briefly.
• Never keep "typing" without either creating the timer or returning a concrete error.

A) Create timer

1. Build a structured intent JSON from the user message (see intent schema reference).
2. Validate it:

printf '%s' "$INTENT_JSON" | bash "{baseDir}/scripts/validate-intent.sh"

1. Upsert timer state:

printf '%s' "$TIMER_JSON" | bash "{baseDir}/scripts/state.sh" upsert

1. Build cron payload(s):

printf '%s' "$TIMER_JSON" | bash "{baseDir}/scripts/build-cron-payload.sh"

1. Call cron tool with action=add for each payload.

Important defaults:

• For reminder output use isolated runs (sessionTarget: "isolated", payload.kind: "agentTurn").
• Route delivery to current channel context unless user requests another target.

B) Scheduled execution (nagging tick)

When a cron reminder turn runs:

1. Call:

bash "{baseDir}/scripts/state.sh" next-nag "<timer_id>"

1. If should_nag=false, do not send a reminder text.
2. If should_nag=true, generate one short reminder using:

- task title - reason (if present) - tone_stage - OpenClaw personality

1. Message must become more urgent as nag_count increases.
2. Apply tone guardrails from escalation policy.

C) User says done

If user explicitly confirms completion (for example "done", "erledigt", "hab ich gemacht"):

1. Resolve active timer.
2. Mark done:

bash "{baseDir}/scripts/state.sh" mark-done "<timer_id>"

1. For recurring timers:

- keep recurring schedule active - reset nagging phase only

D) Ambiguity rules

• If exactly one active nagging timer exists: auto-map completion.
• If multiple active nagging timers exist: ask a short disambiguation question.

E) Fast fallback (important for reliability)

If deterministic helper scripts cannot run for any reason:

• create/update cron jobs directly with the cron tool in the same turn
• use the same schedule and nagging semantics
• confirm the resulting timer setup to the user
• if no schedule can be parsed confidently, ask one short clarification question

Safety and tone constraints

• Keep reminders short (1 to 2 sentences).
• No insults, threats, or manipulative guilt language.
• "More annoyed" means stylistic urgency only, not abuse.
• Never claim the task is done unless user explicitly confirms it.

Packaging note (ClawHub)

Keep this folder self-contained so it can be published directly:

clawhub skill publish ./nervtimer --slug nervtimer --name "NervTimer" --version 0.1.0 --tags latest