Installable skillMIT

Wiki Research: Personal

Weekly research agent for #personal queue items — travel, products, life planning. Runs the 5-step research loop with a practical, actionable focus.

BY v1truv1usUPDATED Oct 27, 2026
INTENDED FOR◈OpenClawInstall in a listed agent.

instructions.md

---
name: wiki-research-personal
description: Weekly research agent for #personal queue items — travel, products, life planning. Runs the 5-step research loop with a practical, actionable focus.
---

# Wiki Research: Personal

Process `#personal` items from the queue — travel, products, life planning, health, and anything not engineering or academic. Uses the 5-step loop tuned for practical decisions.

## Vault location
`/app/data/vault/`

## When to skip the loop
If the topic is a single recommendation you need (a hotel, a specific product), do one `web_search` (engine `brave`) for recent reviews, run `intelli_extract` on the top result, write a one-paragraph page, stop. The full loop is for decisions with trade-offs.

## The 5-step loop

Each step has a gate. Do not advance until the gate is met.

### 1 — Define

Before searching, write three sentences:
- **Assumption:** what you currently believe is true about this topic
- **Decision shape:** the actual decision you're trying to make (buy / choose / plan / schedule / diagnose)
- **Criteria:** what "good" looks like — the 3–5 factors that matter most (price, time, quality, location, durability, etc.)

**Gate:** a reader could list the criteria the decision will be judged on.

### 2 — Triage

Identify the 2–3 source types most likely to answer. For personal topics, priority order:

1. Recent first-hand accounts (reviews, travel reports, build logs) — `web_search` engine `brave`, then `intelli_extract` on individual reviews
2. Comparison / round-up articles from the last 12 months
3. Community discussion (Reddit, Hacker News, specialised forums) for what goes wrong in practice
4. Official specs / datasheets / pricing pages
5. Adjacent vault pages (`wiki/index.md`)

Run 2–3 Brave queries with the criteria from Step 1 in mind. For each promising URL, use `intelli_extract` with a `focusPrompt` naming the specific criterion.

**Gate:** you can name at least one strong source per criterion.

### 3 — Query

Pull the answer from the triaged sources. For personal topics:

- Lead with concrete numbers (price, weight, battery life, time required)
- Capture trade-offs explicitly ("X is better at A but worse at B")
- Capture failure modes ("lasts about 18 months before the hinge cracks")
- Capture recency — for products and travel, anything older than 18 months is suspect

**Gate:** every claim has at least one source URL attached.

### 4 — Verify

For every claim, assign one tier:
- **FACT** — directly stated in a source, with citation
- **LIKELY** — consistent with sources but inferred
- **SPECULATIVE** — your reasoning beyond what sources support
- **UNKNOWN** — neither you nor the sources say

For any claim that drives a money or time decision (price, availability, schedule), run a second `web_search` to corroborate against an independent source.

**Gate:** every claim has a tier.

### 5 — Report

Write the wiki page. Lead with a recommendation or decision matrix if the topic is a choice. Append a Limitations section.

## Wiki page format

```markdown
---
tags: [#personal]
last_researched: YYYY-MM-DD
---

# <Title>

> One-sentence summary.

## Define
- **Decision shape:** ...
- **Criteria:** ...

## Findings

### <Criterion 1>
- **FACT:** <claim> [^1]

### <Criterion 2>
- **FACT:** <claim> [^2]
- **LIKELY:** <claim>

### <Criterion 3>
...

## Recommendation / Trade-offs

If the topic is a choice: state the recommendation with conditions ("X if you prioritise A; Y if you prioritise B").

## Limitations
- What the sources did not cover
- Which claims were not independently verified
- Recency caveats — what's likely to have changed since the sources were written

## Sources
[^1]: <url> — accessed YYYY-MM-DD
[^2]: <url> — accessed YYYY-MM-DD
```

## Per-run workflow

1. Read `RESEARCH_QUEUE.md` in the vault root.
2. Find all unchecked items (`- [ ]`) under `## #personal`. If none, stop.
3. For each real unchecked item, run the 5-step loop.
4. Check `wiki/index.md` for an existing page. Update if found; otherwise create `wiki/personal/<slug>.md`.
5. Update `wiki/index.md`.
6. Mark done in `RESEARCH_QUEUE.md`: `- [x]` + ` — researched YYYY-MM-DD`.
7. Move `[x]` items to `## Archive`.
8. Append entry to `wiki/log.md`:
   ```
   ## [YYYY-MM-DD] ingest | research-queue #personal | <topics> → <pages>
   ```

## Rules
- Never edit `raw/`, `Openclaw articles/`, or `personal stuff/`.
- Only write to `wiki/` and `RESEARCH_QUEUE.md`.
- Lead with concrete numbers and trade-offs, not prose.
- If a topic is too vague to write a Decision shape, mark `[x]` with `— skipped: too vague, clarify topic`.
- Today's date comes from system context.
COMPATIBILITY & REFERENCES

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.