Annotations

Taxonomy note: This skill provides domain expertise (HOW) for processing inline HTML comments for human-agent collaboration. See [[TAXONOMY.md]] for the skill/workflow distinction.

Process inline HTML comments for asynchronous human-agent collaboration across file types.

Supported File Types

Type	Extension	Notes
Markdown	.md	Primary use case. HTML comments invisible in rendered output.
Python	.py	Use # <!-- @ns: comment --> format. HTML comments work but are unconventional.
Other text	.txt, .toml, .yaml, etc.	Any file the agent can read/edit. Use appropriate comment syntax.

Unsupported: Binary files, images, compiled code.

Convention

Human comments: <!-- @nic: comment --> or <!-- @ns: comment --> or <!-- NS: comment --> or '<!-- @nicsuzor: comment -->' Agent responses: <!-- @claude YYYY-MM-DD: response -->

Comments use HTML syntax for universal markdown compatibility (GitHub, Obsidian, VS Code, etc.). They render as invisible in all standard markdown viewers.

Modes

scan [path]

Find all pending human annotations that need response.

<!-- @nicsuzor: use 'rg' instead of grep, run case insensitive, and make the '@' optional --> <!-- @claude 2026-01-12: Done. Updated pattern to use Grep tool (which uses rg), added -i for case insensitive, and made @ optional in the pattern. -->

Grep(pattern="<!--\\s*@?(nic|ns|nicsuzor):", path="$ACA_DATA", output_mode="content", -C=2, -i=true)

Output: List of files with pending comments, showing context.

respond [file]

Process annotations in a specific file:

<!-- @nicsuzor: ensure that comments are added to tasks. Once they are fully captured, they can be deleted from the source file. --> <!-- @claude 2026-01-12: Done. Updated workflow to create tasks for actionable annotations before responding. See step 3 below. --> <!-- @claude 2026-01-24: Updated respond workflow to include step 4: capture annotation threads to bd (backup/documentation) before deletion to ensure full context is preserved. -->

1. Read the file
2. Find all <!-- @nic: or <!-- @ns: comments
3. For each comment without a corresponding <!-- @claude response:
   • Analyze the request in context
   • Add response immediately after: <!-- @claude YYYY-MM-DD: response -->
4. Before deleting resolved annotation threads:
   • Capture the full comment thread (original comment + response) to bd (backup/documentation)
   • Verify the context is preserved for future reference
5. Save the file

clean [file]

Remove resolved annotation threads (both comment and response) after user confirms resolution.

Example

Before:

The court held that platforms must provide notice. <!-- @ns: check if this applies post-DSA -->

After:

The court held that platforms must provide notice. <!-- @ns: check if this applies post-DSA -->

<!-- @claude 2026-01-11: This holding predates DSA. Art. 17 now requires explicit notice with reasoning. Your original cite remains valid for pre-2024 cases. -->

Detection Patterns

Pattern	Matches
<!--\s*@?(nic|ns):	Human comments
<!--\s*@claude\s+\d{4}-\d{2}-\d{2}:	Agent responses
Human comment NOT followed by agent response	Pending items

Workflow Integration

This skill integrates with daily workflow:

1. Morning scan: Run scan $ACA_DATA to find pending annotations
2. Batch respond: Process files with pending comments
3. User review: Human reviews responses, deletes resolved threads

Date Handling

• Agent responses always include date: <!-- @claude 2026-01-11: -->
• Human comments may or may not include dates
• If human comment has date, preserve it; if not, infer from file mtime or mark "undated"

Critical Rules

• ALL matches are actionable: NEVER dismiss matches as "documentation" or "examples". HTML comments are NEVER used casually in this codebase. If the pattern matches, it's a real annotation requiring action.
• Examples use code blocks: Documentation examples appear inside ```markdown code blocks, NOT as bare HTML comments. A bare <!-- @ns: --> in ANY file is a real annotation.
• Direct instructions vs questions: Distinguish between comment types:
  • Questions/discussion (e.g., "check if this applies post-DSA") → add <!-- @claude --> reply
  • Direct instructions (e.g., "re-file this to X folder") → execute the instruction, delete the comment, summarize action in git commit message. Leave files clean.
• Task ID in responses: When creating a task for an annotation, the <!-- @claude --> response MUST include the task ID. Format: <!-- @claude YYYY-MM-DD: Task created: task-id. [brief description] -->

Boundaries

• Scope: Text files in $ACA_DATA and $AOPS that the agent can read and edit
• Supported: Markdown (.md), Python (.py), config files (.toml, .yaml), and other text files
• Never modify: Binary files, compiled code, files outside allowed directories
• Response placement: Immediately after the triggering comment, same line or next line
• Python specifics: Preserve indentation. Place response comment at same indentation level as triggering comment.