gotcha-capture

Capture gotcha learnings from the current conversation and write them into a target skill's ## Gotchas section. This skill mines the conversation for failures, workarounds, edge cases, and non-obvious constraints encountered while using a specific skill, then proposes updates to that skill's SKILL.md.

Step 1: Resolve the target skill

The user provides a skill name as an argument (e.g., /gotcha-capture readme). Resolve it to a SKILL.md file path.

Use the Glob tool to search for the skill:

plugins/*/skills/{argument}/SKILL.md

If no match is found, list all available skills by globbing plugins/*/skills/*/SKILL.md and present the skill names to the user. Use AskUserQuestion to let them pick one.

If the argument matches a plugin name containing multiple skills (e.g., gh contains gh-issue, gh-dev, gh-pr), list all skills inside that plugin and ask the user to specify which one.

If the argument is a file path, use it directly after confirming the file exists.

Store the resolved path as TARGET_SKILL_PATH.

Step 2: Read the target skill

Read the full content of the target SKILL.md using the Read tool.

Parse the file to identify:

• Whether a ## Gotchas section already exists
• If it exists, extract all existing gotcha entries
• Detect the formatting style:
  • Bullet style: - **Bold title** -- Explanation text. (used by readme, gh-pr, gh-issue)
  • Prose style: Paragraph-based entries under the heading

Store the existing gotchas and the detected format style for later use. If no ## Gotchas section exists, default to bullet style.

Step 3: Extract gotchas from conversation

Review the full conversation context for learnings related to the target skill. The conversation is already in your context window — no tools or agents needed for this step.

Look for these patterns:

Errors and failures

• Tool calls that returned errors or unexpected results
• Commands that failed with specific error messages
• Stack traces or exception output
• User statements like "that didn't work", "it failed", "there's an error"

Workarounds

• Sequences where an approach was tried, failed, and a different approach succeeded
• Explicit corrections from the user: "actually, you need to...", "the fix was..."
• Retries with different parameters, flags, or approaches

Unexpected behaviors

• Results that contradicted what the skill's instructions described
• Silent failures where no error occurred but the output was wrong
• Behavior differences across environments or configurations

Non-obvious prerequisites

• Dependencies that needed installing first
• Authentication or permissions issues that blocked progress
• Environment-specific requirements (OS, shell version, tool versions)

Performance issues

• Operations that timed out or were unexpectedly slow
• Resource constraints encountered

Edge cases

• Inputs that caused unexpected behavior (empty repos, large diffs, special characters)
• Boundary conditions or state-dependent issues (detached HEAD, dirty working tree, no remote)

For each gotcha found, draft an entry with:

• A short title (2-5 words) summarizing the issue
• A concise explanation of the problem and how to handle it (1-3 sentences)
• A code snippet if the workaround involves a specific command

If no gotchas are found, inform the user: "I didn't find any gotcha-worthy learnings in this conversation related to {skill-name}. If you have something specific in mind, describe it and I can help draft the entry." Then stop.

Step 4: Merge with existing gotchas

Compare each new candidate against the existing gotcha entries:

• New: The topic is not covered by any existing gotcha. Mark as [NEW].
• Refine: The topic overlaps with an existing gotcha, but the conversation revealed new detail — a better workaround, an additional sub-case, or a more precise description. Mark as [REFINE] and prepare a merged version incorporating both the existing text and the new information.
• Duplicate: The topic is already fully covered by an existing entry. Drop it silently.

If all candidates are duplicates, inform the user: "All the gotchas I found are already documented in {skill-name}'s Gotchas section. No changes needed." Then stop.

Step 5: Present proposed changes

Show the user a clear summary. For each proposed change:

For [NEW] entries:

1. [NEW] **Title** -- Explanation text here.

For [REFINE] entries (show before and after):

2. [REFINE] **Title**
   Before: **Title** -- Original explanation.
   After:  **Title** -- Updated explanation with new detail.

After listing all proposed changes, use AskUserQuestion to confirm. Offer these options:

• Approve all — apply all changes as shown
• Edit entries — let the user revise specific entries before applying
• Cancel — discard everything

If the user asks to edit, present the revised version and confirm again before writing.

Step 6: Write changes to SKILL.md

After user approval, apply the changes using the Edit tool:

If a ## Gotchas section already exists:

• For [NEW] entries: append at the end of the existing Gotchas section, before the next ## heading or end of file.
• For [REFINE] entries: replace the existing entry text with the merged version in its original position.

If no ## Gotchas section exists:

• Insert a new ## Gotchas section. Place it after ## Hard rules if that section exists, otherwise at the end of the file.
• Write all entries under the new heading.

Match the formatting style detected in Step 2. If using bullet style (default), format each entry as:

- **Title** -- Explanation text. `code snippet` if applicable.

After writing, confirm to the user: "Updated {TARGET_SKILL_PATH} with N new gotcha(s) and M refinement(s)."

Hard rules

• Never write to the target SKILL.md without explicit user approval.
• Never delete existing gotcha entries. Only append new ones or refine existing ones.
• Never modify any part of the SKILL.md outside the ## Gotchas section.
• Match the formatting style of the target skill's existing Gotchas section. If none exists, use bullet format with bold titles.
• Each gotcha must describe a real issue encountered in the conversation — never invent hypothetical gotchas.
• Keep gotcha text concise: 2-5 word title, 1-3 sentence explanation.
• When placing a new Gotchas section, insert it after ## Hard rules if present, otherwise at the end of the file. Never insert it in the middle of instructional steps.