| name | history |
| description | Browse and search past analyses from the knowledge system's analysis archive. Use this skill whenever the user invokes `/history` or variants like `/history search=X`, `/history {id}`, `/history --all`, `/history dataset=X`, or asks questions about their analytical work history such as "what have I analyzed before?", "what analyses have I run?", "show me past work", "show my history", "what have we looked at?", "what questions have I answered?", "can I see my analysis history?", "list my analyses", "what did I work on last week?", "have I analyzed X before?", "find analyses about Y", or mentions reviewing prior analyses, checking if similar work was already done, needing context on previous findings, or wants to build on past work. This skill also applies automatically at session start when you need context on the user's analytical history to inform current work. When displaying history, ALWAYS filter to the active dataset unless the user explicitly requests --all or dataset={id}. |
Skill: History
Purpose
Browse and search past analyses from the analysis archive. Helps users
recall what they've analyzed before, find prior findings, and build on
previous work.
When to Use
- User says
/history or "what have I analyzed before?"
- At session start, to provide context on prior work
- When framing a new question, to check if similar analysis exists
Invocation
/history — list recent analyses (last 10)
/history {id} — show full details for a specific analysis
/history search={term} — search by title, question, or tags
/history --all — list all analyses across all datasets
/history dataset={id} — filter to a specific dataset
Instructions
Step 1: Load Archive
- Read
.knowledge/analyses/index.yaml
- If empty: "No analyses archived yet. Complete an analysis and it will appear here."
Step 2: Determine Active Dataset (Critical)
IMPORTANT: Before filtering results, determine the active dataset:
- Read
.knowledge/active.yaml to get the active dataset ID
- If
--all flag is present: skip filtering, show all datasets
- If
dataset={id} is specified: filter to that dataset
- Otherwise: filter to active dataset only
This ensures users see relevant analyses for their current working context by default.
Step 3: Execute Command
List recent (/history):
- Filter to active dataset (unless
--all flag present)
- Sort by date descending
- Show last 10 as a table: date, title, level, key finding count, dataset
- Show total count: "Showing 10 of {total} analyses for {dataset_name}." (or "across all datasets" if --all)
Show specific (/history {id}):
- Find entry by ID in index
- Display: title, date, question, level, all key findings, metrics used,
agents used, output files, tags, confidence, recommendations
- If output files exist, offer: "Want to review the full analysis? The deck is at {path}"
Search (/history search={term}):
- Filter to active dataset first (unless
--all or dataset={id} specified)
- Search across: title, question, key_findings, tags (case-insensitive)
- Display matching entries as a table
- If no matches: "No analyses match '{term}' in {dataset_name}. Try broader terms or use
--all to search across all datasets."
All datasets (/history --all):
- Include dataset_id column in output
- Sort by date descending across all datasets
Step 4: Contextual Suggestions
After displaying history:
- "Want to re-run this analysis with fresh data?"
- "Want to build on finding #{n}?"
- If recent analysis was partial: "This analysis was incomplete. Resume with
/resume-pipeline."
Edge Cases
- No active dataset: Show all analyses or prompt to connect
- Archive file missing: Create empty index
- Analysis output files deleted: Note "output files no longer available"
- Very long history (>100): Paginate, show 20 at a time
Why This Matters
Dataset filtering ensures users see relevant analyses for their current working context. A user analyzing one dataset shouldn't see unrelated-dataset analyses by default — it creates noise and makes it harder to find relevant prior work. The --all flag is available when cross-dataset comparison is needed.
Contextual offers ("Want to review the full analysis?") help users take action on the information. Simply listing file paths isn't enough — explicitly offering next steps makes the history actionable.
Consistent table format allows quick scanning. Users browse history to find patterns, check if work was done before, or locate specific analyses. Tables are faster to scan than prose paragraphs.