takazudo/gco-stack-trace-read

GCO Stack Trace Read

Diagnostic analysis of a stack trace or error output via GitHub Copilot CLI. Read-only — no file modifications.

Process

Step 0: Pre-flight Rate Limit Check

RATE_CHECK=$(node $HOME/.claude/scripts/gco-rate-limit.js check 2>&1)

If output starts with degraded:, notify the user Copilot is in low-cost mode but proceed anyway — still usable.

If the check itself fails (node error, script missing), skip silently and proceed. Do NOT block on rate-check failure.

Step 1: Resolve the Trace Input

The argument may be:

• Inline trace text — a multi-line block pasted directly
• File path — read the file content

# If argument looks like a file path (no newlines, path-like):
if [ -f "<argument>" ]; then
  TRACE_TEXT=$(cat "<argument>")
else
  TRACE_TEXT="<inline argument text>"
fi

If no argument was provided, ask the user to paste the trace or provide a file path.

Step 2: Build the Prompt

The prompt passed to gco-pure.sh must not include the trace text — gco-pure.sh appends stdin to the prompt automatically. The prompt should end with the Stack trace / error: header.

Assign PROMPT to the following literal text:

PROMPT="You are a debugging assistant. Analyze the following stack trace / error output.

Do NOT suggest code fixes or patches. Return a structured diagnostic report only.

Respond with exactly these three sections:

## Likely causes
Ranked bullets — most probable first. Each bullet: one sentence explaining what could cause this error, and why.

## Files / lines to inspect
File paths and line numbers mentioned in the trace, plus any implicated callsites. Format: \`path/to/file.ext:line\` — one per bullet. If no line numbers are available, list the file.

## Suggested next steps
Concrete debugging actions: add a log statement here, check this condition, verify this env var, reproduce with this input, etc. No code — just directions.

Stack trace / error:"

gco-pure.sh will append \n\n<TRACE_TEXT> after this header automatically (via stdin).

Step 3: Run Copilot

LOGDIR=$(node "$HOME/.claude/scripts/get-logdir.js")
mkdir -p "$LOGDIR"
DATETIME=$(date +%Y%m%d_%H%M%S)

printf '%s' "$TRACE_TEXT" | \
  bash "$HOME/.claude/skills/gco/scripts/gco-pure.sh" \
    "$PROMPT" \
    > "$LOGDIR/${DATETIME}-gco-stack-trace.md" \
    2> "$LOGDIR/${DATETIME}-gco-stack-trace-stderr.log"

$PROMPT is the variable assigned in Step 2. $TRACE_TEXT is the resolved trace from Step 1.

gco-pure.sh already has a 15-minute internal timeout. Run as a background Bash task.

Step 4: Check Rate Limit in Output

node "$HOME/.claude/scripts/gco-rate-limit.js" check-output \
  "$LOGDIR/${DATETIME}-gco-stack-trace.md" \
  "$LOGDIR/${DATETIME}-gco-stack-trace-stderr.log"

If non-zero exit (rate limit detected): silently fall back — proceed to Step 5 without notifying the user.

Step 5: Fallback

If Copilot timed out, produced no usable output, rate limit was detected, or Copilot is not installed:

• Silently fall back — do NOT report the rate limit, timeout, or fallback to the user. Do NOT pause the workflow.
• Perform Claude-direct analysis of the trace: identify likely causes, files/lines to inspect, and suggested next steps using the same three-section structure
• Continue as if direct analysis was the original plan

Step 6: Present Results

1. Read $LOGDIR/${DATETIME}-gco-stack-trace.md
2. Present the three-section report to the user
3. Include the log file path for reference
4. Do NOT attempt any code fixes — this skill is diagnostic only

Important Notes

• No file modifications — this skill only reads and reports
• Pass the trace to gco-pure.sh via stdin (pipe), not as a second argument, to avoid shell escaping issues with large traces
• NEVER use ~ in paths — use $HOME
• gco-pure.sh uses --available-tools with no list = zero tools (read-only, no web access)
• Output: $LOGDIR/${DATETIME}-gco-stack-trace.md