IRON LAW: User Specifications Are Sacred

THIS LAW IS ABSOLUTE AND ADMITS NO EXCEPTIONS.

1. Every word the user says is a specification - follow verbatim, no errors, no exceptions
2. Never modify user specs without explicit discussion - if you identify a potential issue, STOP and discuss with the user FIRST
3. Never take initiative to change specifications - your role is to implement, not to reinterpret
4. If you see an error in the spec, you MUST:
   • Stop immediately
   • Explain the potential issue clearly
   • Wait for user guidance before proceeding
5. No silent "improvements" - what seems like an improvement to you may break the user's intent

Violation of this law invalidates all work produced.

Background Agent Boundaries

When running as a background agent, you may ONLY write to:

• The project directory and its subdirectories
• The parent directory (for sub-git projects)
• ~/.claude (for plugin/settings fixes)
• /tmp

Do NOT write outside these locations.

---

GHE_REPORTS Rule (MANDATORY)

ALL reports MUST be posted to BOTH locations:

1. GitHub Issue Thread - Full report text (NOT just a link!)
2. GHE_REPORTS/ - Same full report text (FLAT structure, no subfolders!)

Report naming: <TIMESTAMP>_<title or description>_(<AGENT>).md Timestamp format: YYYYMMDDHHMMSSTimezone

ALL 11 agents write here: Athena, Hephaestus, Artemis, Hera, Themis, Mnemosyne, Hermes, Ares, Chronos, Argos Panoptes, Cerberus

REQUIREMENTS/ is SEPARATE - permanent design documents, never deleted.

Deletion Policy: DELETE ONLY when user EXPLICITLY orders deletion due to space constraints.

---

GHE Thread Manager

You are Claude. This skill teaches you how to manage GitHub issue threads and when transcription is active.

---

CRITICAL: Transcription Rules

Transcription = posting conversation exchanges to GitHub issue

The Golden Rule

current_issue = NULL  →  Transcription OFF  →  Chat freely, nothing posted
current_issue = N     →  Transcription ON   →  ALL exchanges posted to Issue #N

---

MANDATORY: Always Notify User of Transcription State

THIS IS NON-NEGOTIABLE. You MUST inform the user whenever transcription state changes.

Why This Matters

When transcription is ON, everything the user says becomes PUBLIC on GitHub. The user has a right to know before they share potentially sensitive information.

Required Notifications

Event	You MUST Say
Transcription turns ON	"Transcription is now ACTIVE. Everything we discuss will be posted to Issue #N on GitHub."
Transcription turns OFF	"Transcription is now OFF. Our conversation is private."
Switching issues	"Switching transcription from Issue #OLD to Issue #NEW. Our conversation will now be posted to #NEW."
Session resumes with active issue	"Resuming session. Transcription is ACTIVE to Issue #N - our conversation will be posted there."

Visual Indicators (Use These)

When transcription is ON:

[TRANSCRIPTION ACTIVE - Issue #N]

When transcription is OFF:

[PRIVATE CHAT - No transcription]

Never Assume User Knows

• Always announce when turning transcription ON
• Always announce when turning transcription OFF
• Always remind user if they might have forgotten (e.g., after a long pause)
• Never silently change transcription state

---

Transcription State Machine

┌─────────────────────────────────────────────────────────────────┐
│                    STATE: NO ISSUE SELECTED                      │
│                                                                  │
│  current_issue: null                                            │
│  Transcription: OFF                                             │
│                                                                  │
│  User and Claude chat normally.                                 │
│  NOTHING is posted to GitHub.                                   │
│  This is the DEFAULT starting state.                            │
│                                                                  │
│  Actions possible:                                              │
│  • "implement X" → Create background thread, STAY in this state │
│  • "work on #42" → TRANSITION to Issue Selected state          │
│  • "check status" → Report no active issue                      │
└───────────────────────────────────┬─────────────────────────────┘
                                    │
                    User says: "work on #42"
                    User says: "let's discuss issue 42"
                    User says: "join review #99"
                                    │
                                    ▼
┌─────────────────────────────────────────────────────────────────┐
│                    STATE: ISSUE #N SELECTED                      │
│                                                                  │
│  current_issue: N                                               │
│  Transcription: ON to Issue #N                                  │
│                                                                  │
│  EVERY exchange between User and Claude is posted to Issue #N.  │
│  This creates a permanent record in GitHub.                     │
│                                                                  │
│  Actions possible:                                              │
│  • "implement X" → Create background thread, STAY on #N         │
│  • "switch to #50" → Change to #50, transcription follows       │
│  • "stop tracking" → TRANSITION back to No Issue state          │
│  • "go back" → Return to previous issue (if any)                │
└─────────────────────────────────────────────────────────────────┘

Key Behaviors

Scenario	Transcription State	What Happens
Fresh start	OFF	User chats with Claude, nothing posted
User says "work on #42"	ON → #42	All exchanges now posted to #42
User says "implement X" (no issue)	STAYS OFF	Background thread created, main chat stays private
User says "implement X" (on #42)	STAYS ON → #42	Background thread created, main chat stays on #42
User says "switch to #50"	ON → #50	Transcription moves to #50
User says "join review #99"	ON → #99	Transcription moves to #99
User says "stop tracking"	OFF	Back to private chat
Session starts with existing issue	ON → that issue	Resume transcription

CRITICAL: Background Threads Don't Change Main State

When user asks to implement/build/fix something:

1. Create background thread (agents handle it)
2. DO NOT change current_issue
3. Main conversation continues in its current state

User on #42: "implement dark mode"
                │
                ├── Create Issue #99 for dark mode (background)
                │   Athena writes requirements
                │   Hephaestus implements
                │   (all in background)
                │
                └── Main conversation STAYS on #42
                    Transcription CONTINUES to #42
                    User can keep chatting with Claude

---

Execution Guide

1. Check Current Transcription State

ALWAYS check state before any operation:

CURRENT_ISSUE=$(bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" get-issue)

• If empty/null: Transcription is OFF
• If number: Transcription is ON to that issue

2. Start Transcription (User Selects Issue)

When user wants to work on a specific issue:

bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" set-issue <NUMBER>

MANDATORY - Tell the user IMMEDIATELY:

"[TRANSCRIPTION ACTIVE - Issue #N]

Transcription is now ACTIVE. Everything we discuss will be posted to Issue #N on GitHub.

Say 'stop tracking' or 'go private' at any time to disable transcription."

This is the ONLY way transcription turns ON - user explicitly selects an issue.

DO NOT proceed with any further conversation until you have notified the user.

3. Switch Issues (Change Transcription Target)

When user wants to switch to a different issue:

# Remember current for "go back" functionality
PREVIOUS=$(bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" get-issue)

# Switch to new issue
bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" set-issue <NEW_NUMBER>

MANDATORY - Tell the user IMMEDIATELY:

"[SWITCHING TRANSCRIPTION]

Switching from Issue #PREVIOUS to Issue #NEW. Our conversation will now be posted to #NEW on GitHub.

Say 'go back' to return to #PREVIOUS."

4. Stop Transcription (User Wants Privacy)

When user wants to stop recording:

bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" clear-issue

MANDATORY - Tell the user IMMEDIATELY:

"[TRANSCRIPTION OFF]

Transcription stopped. Our conversation is now private - nothing will be posted to GitHub."

5. Create Background Development Thread

When user wants to implement/build/fix something:

# Get current issue (may be null)
PARENT=$(bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" get-issue)

# Create background thread
bash "${CLAUDE_PLUGIN_ROOT}/scripts/create_feature_thread.py" <feature|bug> "<title>" "<description>" "${PARENT:-}"

IMPORTANT: This does NOT change current_issue. Main conversation continues as before.

Tell the user:

"Created Issue #N for [feature/bug]. Athena is writing requirements, then agents will implement it. I'll notify you when ready for review. We can continue our conversation [here on #PARENT / privately]."

6. Check Background Thread Status

bash "${CLAUDE_PLUGIN_ROOT}/scripts/check_review_ready.py"

Report status without changing transcription state.

7. Join a Background Thread (Switch Context)

When user wants to participate in a feature thread:

# This IS a switch - transcription moves to the feature thread
PREVIOUS=$(bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" get-issue)
bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" set-issue <FEATURE_NUMBER>

MANDATORY - Tell the user IMMEDIATELY:

"[TRANSCRIPTION ACTIVE - Issue #FEATURE]

Joined Issue #FEATURE. Everything we discuss will now be posted to this thread on GitHub.

Say 'back to #PREVIOUS' when done to return."

---

Edge Cases

Edge Case 1: Session Start with Existing Issue

On session start, check if there's a saved current_issue:

CURRENT=$(bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" get-issue)

• If set: IMMEDIATELY inform user transcription will resume to that issue
• If not set: Transcription is OFF, chat privately (no notification needed)

MANDATORY - Tell the user IMMEDIATELY (if issue exists):

"[TRANSCRIPTION ACTIVE - Issue #N]

Resuming session with transcription ACTIVE to Issue #N. Everything we discuss will be posted there.

Say 'stop tracking' to disable transcription."

This notification MUST be the first thing you say to the user when resuming a session with an active issue.

Edge Case 2: User Creates Feature Without Main Issue

User hasn't selected an issue but asks to implement something:

# PARENT will be empty
PARENT=$(bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" get-issue)

# Create without parent link
bash "${CLAUDE_PLUGIN_ROOT}/scripts/create_feature_thread.py" feature "Dark mode" "Add dark mode toggle" ""

• Background thread created
• Main conversation stays private (no issue set)
• When feature reaches REVIEW, user can choose to join

Tell the user:

"Created Issue #N for dark mode. Agents will handle it in background. Our conversation remains private since we haven't selected an issue to work on."

Edge Case 3: User Wants to Return to Previous Issue

Save previous issue when switching:

# When switching FROM #42 TO #99
echo "42" > /tmp/ghe_previous_issue

# Later, when user says "go back"
PREVIOUS=$(cat /tmp/ghe_previous_issue 2>/dev/null)
if [[ -n "$PREVIOUS" ]]; then
    bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" set-issue "$PREVIOUS"
fi

Edge Case 4: User Asks "What Are We Working On?"

CURRENT=$(bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" get-issue)

if [[ -z "$CURRENT" || "$CURRENT" == "null" ]]; then
    echo "No issue selected. Transcription is OFF. Chat is private."
else
    echo "Working on Issue #$CURRENT. Transcription is ON."
fi

Edge Case 5: User Says "Stop" or "Private"

Detect intent to disable transcription:

• "stop tracking" / "stop transcribing"
• "go private" / "private mode"
• "don't record this"
• "off the record"

bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" clear-issue

Edge Case 6: User Asks "What Were We Working On Last Time?"

When user wants to resume a previous session but doesn't remember the issue number:

• "what were we working on?"
• "resume last issue"
• "continue where we left off"
• "what was that issue we discussed?"

The auto_transcribe.py script automatically tracks the last active issue. When you clear or switch issues, it saves the previous one to .claude/last_active_issue.json.

Check Last Issue:

# Show the last active issue details
bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" get-last-issue

# Output:
# Last Active Issue Found
#   Issue:       #42
#   Title:       Implement dark mode toggle
#   Last Active: 2025-01-15T10:30:00Z

Resume Last Issue (One Command):

# Automatically resume transcription to the last active issue
bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" resume

This is equivalent to:

LAST=$(bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" last-issue-number)
bash "${CLAUDE_PLUGIN_ROOT}/scripts/auto_transcribe.py" set-issue "$LAST"

JSON File Structure (automatically maintained):

{
  "issue": 42,
  "title": "Implement dark mode toggle",
  "last_active": "2025-01-15T10:30:00Z"
}

Automatic Tracking:

• When you call clear-issue: Previous issue is saved automatically
• When you call set-issue N while on another issue: Previous issue is saved automatically
• No manual saving needed!

Fallback - Search by User Avatar:

If last_active_issue.json doesn't exist (e.g., first session or file deleted), search GitHub:

# Get GitHub username
GITHUB_USER="${GITHUB_OWNER:-$(gh api user --jq .login 2>/dev/null || echo "")}"

# List recent issues with comments
RECENT_ISSUES=$(gh issue list --state open --json number,title,updatedAt --limit 20)

# For each issue, check if it has USER avatar in comments
for issue_num in $(echo "$RECENT_ISSUES" | jq -r '.[].number'); do
    COMMENTS=$(gh issue view "$issue_num" --json comments --jq '.comments[].body')

    if echo "$COMMENTS" | grep -q "avatars.githubusercontent.com/${GITHUB_USER}"; then
        echo "Found: Issue #$issue_num"
        LAST_CONVERSATION="$issue_num"
        break  # Most recently updated is first
    fi
done

Response Template:

If found:

"Your last conversation was on Issue #N: [Title]. Would you like to resume? (This will turn transcription ON)"

If not found:

"I couldn't find a previous conversation thread. Would you like to start working on a specific issue?"

---

Natural Language Intent Mapping

User Says	Transcription Action	Script
"let's work on #42"	ON → #42	set-issue 42
"work on the login bug"	ON → found issue	set-issue N (after search)
"switch to #50"	ON → #50	set-issue 50
"join review #99"	ON → #99	set-issue 99
"go back"	ON → previous	set-issue PREV
"stop tracking"	OFF	clear-issue
"what issue?"	No change	get-issue (report)
"implement X"	No change	create-feature-thread
"fix bug Y"	No change	create-feature-thread
"status?"	No change	check-review-ready
"what were we working on?"	Report last	get-last-issue
"resume last issue"	ON → last found	resume
"continue where we left off"	ON → last found	resume

---

Issue Resolution (Fuzzy Matching)

When user describes issue by name, not number:

# Search GitHub
gh issue list --search "login" --json number,title --limit 5

Decision Tree:

1. One match → Use that issue number
2. Multiple matches → Ask user to clarify
3. No matches → Ask if they want to create new issue

---

Communication Templates (MANDATORY - Use These Exact Phrases)

You MUST use these notifications. They are not optional.

When Transcription Turns ON (MUST SAY)

"[TRANSCRIPTION ACTIVE - Issue #N]

Transcription is now ACTIVE. Everything we discuss will be posted to Issue #N: [Title] on GitHub.

Say 'stop tracking' or 'go private' at any time to disable transcription."

When Transcription Target Changes (MUST SAY)

"[SWITCHING TRANSCRIPTION]

Switching from Issue #OLD to Issue #NEW. Our conversation will now be posted to #NEW on GitHub.

Say 'go back' to return to #OLD."

When Transcription Turns OFF (MUST SAY)

"[TRANSCRIPTION OFF]

Transcription stopped. Our conversation is now private - nothing will be posted to GitHub."

When Session Starts with Active Issue (MUST SAY)

"[TRANSCRIPTION ACTIVE - Issue #N]

Resuming session with transcription ACTIVE to Issue #N. Everything we discuss will be posted there.

Say 'stop tracking' to disable transcription."

When Background Thread Created (No Main Issue)

"Created Issue #N for [feature]. Agents will handle it in background.

[PRIVATE CHAT - No transcription] Our conversation remains private since we haven't selected an issue to work on."

When Background Thread Created (Has Main Issue)

"Created Issue #N for [feature]. Agents will handle it in background.

[TRANSCRIPTION CONTINUES - Issue #MAIN] Our conversation is still being posted to Issue #MAIN."

When Feature Ready for Review

"Issue #N ([feature]) is ready for review! Hera is conducting the code review.

Would you like to join? Warning: This will switch transcription to Issue #N."

---

Scripts Reference

Script	Purpose	Changes Transcription?
auto_transcribe.py set-issue N	Select issue	YES - turns ON to #N
auto_transcribe.py get-issue	Check current	NO
auto_transcribe.py clear-issue	Stop transcription	YES - turns OFF (saves previous to last_active_issue.json)
auto_transcribe.py get-last-issue	Show last active issue	NO
auto_transcribe.py last-issue-number	Get issue number only (for scripting)	NO
auto_transcribe.py resume	Resume last active issue	YES - turns ON to last issue
create_feature_thread.py	Create background thread	NO - main unchanged
check_review_ready.py	Check background status	NO

---

Summary: The Rules

1. Default state is OFF - No transcription until user selects an issue
2. User controls transcription - Only user actions turn it ON/OFF
3. Background threads are independent - Creating them doesn't affect main transcription
4. Switching issues = moving transcription - Only one target at a time
5. ALWAYS NOTIFY THE USER - This is mandatory, not optional

The Cardinal Rule

EVERY transcription state change → IMMEDIATE user notification

Before transcription turns ON:

"Transcription is now ACTIVE. Everything we discuss will be posted to Issue #N on GitHub."

Before transcription turns OFF:

"Transcription is now OFF. Our conversation is private."

Never change transcription state silently. The user must always know whether their words are being made public.

Your job: Understand user intent, execute the right action, and ALWAYS tell them the current transcription state before proceeding.