/explore

Read ## Vault Structure from ~/.claude/vault-local.md (fall back to ~/.claude/CLAUDE.md if not found) to get the vault path, the research folder location, and the list of content folders. If the section does not exist or required fields are missing, stop and respond: "Setup not complete. Please run /hirameki:__init first."

Analyse the input and automatically select a mode based on its shape.

Input: $ARGUMENTS

• If $ARGUMENTS is empty, ask: "Explore what? (concept, two topics, question, or test: argument)" and wait for the answer before continuing. Ask in the language specified in ## Vault Structure → language.

Mode detection

Input shape	Mode	Example
Single concept or word	Arc — concept evolution timeline	explore agency
Two topics separated by and, 與, or と	Bridge — hidden connections between two topics	explore design and engineering
Ends with ? or ？	Ghost — answer in your voice	explore how much autonomy should AI have?
Starts with test:	Stress-test — pressure-test the argument	explore test: the limits of prompt engineering

Append save to any input to write the result to a file.

---

Arc mode — concept evolution tracking

Scan scope: entire vault, prioritising daily and content folders.

Logic:

1. Check {research}/arc/ for an existing file about the same concept today (match by filename keyword)
2. If found → append mode
3. If a possibly related file is found but uncertain → list candidates and ask whether to append or create new
4. If not found → create mode

Output structure:

# Concept tracking: {concept}

> Analysis time: YYYY-MM-DD HH:MM

## First appearance
Earliest file where this concept appeared, and the context. Quote up to 3 sentences.
Date: YYYY-MM-DD

## Evolution timeline
- YYYY-MM-DD | [[filename]] | How this concept was used or positioned in this file (one line)

## Current state
Which topics this concept connects to now, any contradictory uses, any drafts developing it.

## Unexplored angles
Aspects of this concept not yet addressed anywhere in the vault.

Append mode adds to the existing file:

---

## Tracking update [HH:MM]

### Timeline additions
- [New mentions since last analysis]

### State changes
- [What changed since last analysis. If nothing, write "No significant change"]

### Unexplored angles (updated)
- [Re-evaluate what remains unexplored]

Write target (when saving): {research}/arc/YYYY-MM-DD-{concept}.md

---

Bridge mode — hidden connections between two topics

Scan scope: entire vault.

Logic:

1. Check {research}/bridge/ for an existing file about the same pair today (match by keywords, order-independent)
2. If found → append mode; if not → create mode

Output structure:

# Bridge analysis: {topic A} and {topic B}

> Analysis time: YYYY-MM-DD HH:MM

## Direct intersections
Files that mention both topics. For each: [[filename]], how the two topics are related there.
If none, write "No direct intersections."

## Bridge notes
Files that mention only one topic but could form a connection. Explain why each might be a bridge.
Limit: 5 files.

## Connection hypotheses
1–3 hypotheses about deeper connections between the two topics.
For each: hypothesis, confidence level (strong / medium / weak), evidence.

Append mode adds to the existing file:

---

## Tracking update [HH:MM]

### New intersections
- [New intersections or bridge notes since last analysis]

### Hypothesis validation
- [New evidence supporting or contradicting earlier hypotheses]

Write target (when saving): {research}/bridge/YYYY-MM-DD-{topic-A}-and-{topic-B}.md

---

Ghost mode — answer in your voice

Scan scope: all completed articles in content folders (exclude drafts/ and thoughts/ subdirectories).

Steps:

1. Analyse writing style: sentence patterns, vocabulary, argument structure, recurring rhetorical moves
2. Extract existing positions and arguments relevant to the question
3. Compose an answer using that style and those positions

Output structure:

[Answer]
Answer the question in the author's voice. Match the length of a typical paragraph from their writing.

[Evidence]
List the notes referenced: [[filename]], the specific passage or position drawn from. Limit: 5.

[Confidence annotation]
Note which parts of the answer have direct vault support, and which are style-inferred extensions.

Write target (when saving): {research}/ghost/YYYY-MM-DD-{question-summary}.md

---

Stress-test mode — pressure-test the argument

Input: strip the leading test: — the remainder is the topic.

Scan scope: all files in the vault that mention the topic.

Steps:

1. Collect all claims and arguments about the topic from the vault
2. Examine the logical structure and evidence base of each claim

Output structure:

[Claim list]
All major claims about this topic found in the vault. For each: claim (one sentence), source [[filename]].

[Weakness analysis]
For each claim, check the following (only list what applies):
- Internal contradiction: conflicting statements across vault files
- Unverified assumption: claim rests on an unproven premise
- Logic gap: missing steps in the argument
- Evidence gap: claim lacks supporting data or examples
Cite specific [[filename]] and passage for each weakness.

[Overall assessment]
How solid is the overall argument for this topic: solid / mostly solid with gaps / needs major work.
List the 1–3 weaknesses most worth addressing first.

Write target (when saving): {research}/stress-test/YYYY-MM-DD-{topic-summary}.md

---

Shared rules

• If search results exceed 20 files, list only the 20 most relevant and note the total count
• Use [[wiki link]] format for all file references
• Timestamps use local time in HH:MM format (24-hour)
• Always print the full analysis output to the terminal
• Show the filename, mode (create / append), content summary, and full path, then wait for confirmation before writing
• Print the full path after writing
• Create the folder if it does not exist, and print the full path

Write output in the language specified in ## Vault Structure → language.