菜单
Give Claude a sense of time -- tracks elapsed time between messages, task durations, project switching, and activity patterns. Hooks inject temporal context into every prompt automatically. Includes a portable Python module for wrapping any LLM call with time awareness.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use after completing a major work batch (plan execution, feature build, multi-task subagent run) to refresh code-graph-derived data in MEMORY.md/CONTEXT.md/SKILL_MEMORY, append dated narrative entries to TODO/CHANGELOG/lessons-learned, mine cross-project structural habits, and ratchet reusable permissions to global settings. Keeps project DNS in sync with reality.
Use when orchestrating multiple ghengis-skills across a single task -- coordinates pre-action validation, execution, post-action verification, and lesson extraction via a shared JSON scratchpad. TRIGGER when dispatching a subagent, finishing a major task, or when the user mentions "run the full chain", "validate this", "fire the pipeline", or explicitly invokes a chain by name.
Use to build and maintain an evolving understanding of user preferences, communication style, and working patterns — learns from interactions and adapts behavior over time
Use when starting a new project or organizing an existing one — creates self-documenting 4-layer project structure with MEMORY.md, CONTEXT.md, modular CLAUDE.md, and workspace routing
Use to accumulate and retrieve domain knowledge from past tasks — builds a grepable plain-text knowledge base that grows over time, enabling agents to learn from experience without vector databases
Use when reorganizing an existing project's file tree based on what actually connects to what — analyzes import graph, classifies layers, proposes an ideal target tree, validates the plan adversarially, then executes file moves with atomic per-language import rewrites. Always plan-first; never moves a file before the user confirms. TRIGGER on "reorganize this project", "clean up the file tree", "the structure is a mess", "make this match a real project layout", or after project-scaffold detects layout drift.
| name | time-perception |
| description | Give Claude a sense of time -- tracks elapsed time between messages, task durations, project switching, and activity patterns. Hooks inject temporal context into every prompt automatically. Includes a portable Python module for wrapping any LLM call with time awareness. |
| allowed-tools | Read Bash |
A time-awareness system that gives Claude (or any LLM) a persistent sense of time across conversations. Hooks inject a compact timestamp into every prompt, and log files accumulate history for pattern analysis.
Every time a user sends a message, a UserPromptSubmit hook runs a shell script that:
A Stop hook runs when Claude finishes responding, logging how long the task took.
The injected string looks like:
[T:2026-04-19 01:14|+29m 14s|#725]
Fields:
T: -- current local datetime+ -- elapsed time since previous message# -- cumulative message number across all sessionsOn project switch, an extra field appears:
[T:2026-04-19 01:14|+5m 2s|#725|Jarvis<-General Claude]
When you see the time context injected into a conversation:
If the elapsed time is > 1 hour, briefly acknowledge the return:
The Stop hook logs how long each response took. Use this data when the user asks about productivity:
The JSONL logs accumulate over days/weeks. When the user asks about their habits:
When switched_from_project appears, you know the user just came from another repo. This is useful context:
All stored under ~/.claude/ (or a custom data_dir):
| File | Purpose |
|---|---|
time-data.json | Current state (last timestamp, message count, last project) |
time-log.jsonl | Global message history (timestamp, elapsed, project, agent) |
task-durations.jsonl | Global task duration log (how long each response took) |
project-time/<project>.jsonl | Per-project message log with prompt previews |
project-time/<project>.durations.jsonl | Per-project task duration log |
{
"timestamp": "2026-04-19T01:14:00Z",
"local": "2026-04-19 01:14:00 MST",
"epoch": 1776489240,
"elapsed_seconds": 1754,
"message_number": 725,
"project": "C:/Users/user/projects/my-app"
}
{
"completed": "2026-04-19 01:15:23 MST",
"duration_seconds": 83,
"duration_human": "1m 23s",
"project": "my-app"
}
Add these to your settings.json (user-level or project-level):
{
"hooks": {
"UserPromptSubmit": [
{
"hooks": [
{
"type": "command",
"command": "bash ~/.claude/scripts/time-tracker.sh"
}
]
}
],
"Stop": [
{
"hooks": [
{
"type": "command",
"command": "bash ~/.claude/scripts/task-timer.sh"
}
]
}
]
}
}
Copy time-tracker.sh and task-timer.sh from this skill's scripts/ directory to ~/.claude/scripts/.
Copy time_context.py to ~/.claude/scripts/ for use in custom scripts, agents, or LLM wrappers.
time_context.py is a standalone module with no dependencies beyond the standard library. It can be imported by any Python script, LLM wrapper, or subagent.
from time_context import TimeContext
tc = TimeContext(project="my-app", agent_name="gpt-4")
# Get time context string to inject into any prompt
context = tc.stamp("prompt preview text")
# -> "[T:2026-04-19 01:14|+5m 2s|#42]"
# Mark task complete (logs duration)
duration = tc.done("task label")
# -> "1m 23s"
# Wrap any LLM call with automatic time tracking
response = tc.wrap(my_llm_function, "fix the bug in auth.py")
# Inject into OpenAI-style messages list
messages = tc.inject([{"role": "user", "content": "hello"}])
# -> [{"role": "system", "content": "[T:...]"}, {"role": "user", "content": "hello"}]
# Get project summary for analysis
summary = tc.summary()
# -> {"project": "my-app", "total_messages": 42, "avg_task_duration": "45s", ...}
# Read raw history
history = tc.get_history(project_only=True)
durations = tc.get_durations(project_only=True)
TimeContext(
project: str = None, # Project name (auto-detected from cwd if None)
data_dir: str = None, # Storage directory (default: ~/.claude/)
agent_name: str = None, # Agent identifier for multi-agent tracking
)
| Method | Returns | Purpose |
|---|---|---|
stamp(prompt_preview) | str | Record timestamp, return context string |
done(task_label) | str | None | Log task duration, return human-readable time |
wrap(llm_call, prompt) | Any | Wrap an LLM call with automatic time tracking |
inject(messages) | list[dict] | Prepend time context as system message |
get_history(project_only) | list[dict] | Read message history from logs |
get_durations(project_only) | list[dict] | Read task duration history |
summary() | dict | Generate project time summary |
Add this to your global ~/.claude/CLAUDE.md:
## Time Awareness
A `UserPromptSubmit` hook injects time context into every conversation. Always acknowledge and use this temporal data:
- Note how long it's been since the user's last message
- When time gaps are large (>1 hour), briefly acknowledge the return
- Data files for analysis:
- `~/.claude/task-durations.jsonl` -- global task duration log
- `~/.claude/time-log.jsonl` -- global message timestamp history
- `~/.claude/project-time/<project>.jsonl` -- per-project message log
- `~/.claude/project-time/<project>.durations.jsonl` -- per-project task duration log
[T:...|+...|#...] format is ~40 bytes. Full XML would waste tokens.