Skill

mnemo-timeline

Render the temporal/decision shape of a past session - phases, gaps, tool bursts, compact boundary, broken-prompt candidates. Use when the user asks how a session unfolded, what the decision arc was, where a decision happened, or when reconstructing a session after compacting.

npx claudepluginhub kawaiillm/claude-mnemo --plugin claude-mnemo

Tool Access

This skill uses the workspace's default tool permissions.

Preview

Timeline is the temporal axis of the three-axis read model. It answers "how did this session unfold" by rendering a time-ordered turn table, session-wide phase segmentation, and window-scoped shape signals.

SKILL.md

Similar Skills

cache-components

139.3k

Guides Next.js Cache Components and Partial Prerendering (PPR): 'use cache' directives, cacheLife(), cacheTag(), revalidateTag() for caching, invalidation, static/dynamic optimization. Auto-activates on cacheComponents: true.

cache-components

mcp-builder

124.2k

Guides building MCP servers enabling LLMs to interact with external services via tools. Covers best practices, TypeScript/Node (MCP SDK), Python (FastMCP).

9 files

anthropics-skills-13

Stats

Parent Repo Stars0

Parent Repo Forks0

Last CommitApr 12, 2026

Actions

View Source View Plugin View on GitHub View README

Help us improve

Share bugs, ideas, or general feedback.

Mnemo Timeline

Three axes of read access:

recall - content axis: structured semantic index
timeline - temporal axis: decision arc, phases, gaps, bursts
mnemo-replay - raw axis: direct JSONL + SQLite access

remember is the single write path. Use timeline for shape, recall for content, and mnemo-replay only for exact bytes.

When to use

Use timeline when:

Reconstructing a session after compacting
Asking how a session unfolded or where the decision arc changed
Finding where a decision was made
Debugging an in-progress session with pending turns
Inspecting gaps, tool bursts, broken-prompt candidates, or compact boundaries

Input

timeline(id="S42")

The input has exactly one field: id. No depth, page, pageSize, query, or truncate.

Range syntax

Form	Returns
`S42`	First 30 turns, or the whole session if shorter
`S42/T*`	Same as `S42`
`S42/T10..30`	Closed range `T10-T30`
`S42/T10..50`	Truncated to 30 rows
`S42/T..20`	Open-start range `T1-T20`
`S42/T30..`	Open-end range starting at `T30`

Hard cap: timeline returns at most 30 turns in the table. Phases and session metadata remain session-wide.

timeline(id="S42/T10") is an error. Use recall(id="S42/T10", depth="expanded") for single-turn detail.

Output structure

Session header

The header includes:

Session id, local-time range, and duration
Session stats and type counts
showing: line with the current window and next hint
Explicit timezone line with abbreviation and UTC offset
raw: line with the absolute JSONL path for mnemo-replay

Turn table

The turn table has two content columns: prompt and title.

prompt column: cleaned raw user prompt, capped at 200 chars
title column: <type_emoji> <Mnemosyne title> when extracted, ⏳ when pending, strikethrough when undone

Markers:

⨯ prefix on T# for undone turns
※ in the gap column for broken-prompt candidates
[ext:<name>] prefix in the prompt column for external-source turns

Phases

Phases are run-length encoded by turn type across the whole session. Labels use the turn.type enum values. Pending turns form their own pending phase.

Shape signals

Shape signals are computed over the returned window only and include:

Fastest and longest gaps
Tool bursts
Broken-prompt candidates
Undone turns
External inputs
Compact boundary

Workflow examples

recall()
timeline(id="S42")
recall(id="S42/T19", depth="expanded")
# Raw: use mnemo-replay on the path from the timeline header

Design notes

Timeline reads only SQLite
Timeline is main-agent only
Cross-session timelines are out of scope for v1
Local-timezone rendering uses Intl.DateTimeFormat
The 30-turn cap is intentional