Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/backnotprop/plannotator/annotate-lastgit clone --depth 1 https://github.com/backnotprop/plannotatorWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00018 | $0.00679 |
| Opus 5 | $0.00009 | $0.00340 |
| Sonnet 5 | $0.00004 | $0.00136 |
| Haiku 4.5 | $0.00002 | $0.00068 |
Grade A, and why
annotate-last scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 95 lines — stays where its author put it; the contents beside it link to each section on GitHub.
The /plannotator-last command opens the agent's most recent response in the annotation UI, letting you highlight text, add comments, and send structured feedback back.
Usage
Claude Code
/plannotator-last
OpenCode
/plannotator-last
Pi
/plannotator-last
Codex
!plannotator last
How it works
User runs /plannotator-last
↓
Last assistant message extracted from session
↓
Annotate server starts (random port)
↓
Browser opens, loads annotation UI
↓
/api/plan returns { plan: message, mode: "annotate-last" }
↓
User annotates → Send Annotations
↓
Feedback sent to agent
Session log parsing
Each harness reads the last assistant message differently:
| Harness | Source | Method |
|---|---|---|
| Claude Code | ~/.claude/projects/{slug}/*.jsonl |
Parses JSONL session logs, finds last assistant text blocks |
| OpenCode | SDK | client.session.messages() API |
| Pi | SDK | ctx.sessionManager.getEntries() API |
| Codex | ~/.codex/sessions/ rollout files |
Parses JSONL by CODEX_THREAD_ID env var |
For Claude Code, the parser handles streamed chunks (multiple JSONL lines sharing the same message.id), filters out system-generated user messages, and skips noise entries. If the most recent session log has no assistant messages, it tries earlier logs sorted by modification time.
Annotate-last mode differences
The annotation UI in annotate-last mode works the same as /plannotator-annotate, with minor copy changes:
- Copy button shows "Copy message" instead of "Copy plan"
- Completion screen says "annotations on the message"
- Feedback export is titled "Message Feedback" instead of "Plan Feedback"
Flags
plannotator annotate-last accepts the same --gate, --json, and --hook flags as plannotator annotate. See Annotate → Flags for the full matrix.
The common use case for --gate on annotate-last is a turn-by-turn review gate wired to a Stop hook:
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 95 lines · 18 tokens per session scan A 0c6a3a43d347
annotate-last is a command published in the GitHub repository backnotprop/plannotator (8,256 stars, last pushed 2d ago), licensed Apache-2.0. It adds 18 tokens to every session and 679 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other commands, from other repositories
notebooklm
Vault-first source-grounded research via Gemini File Search. One command, no browser. The grounded parallel to /research-deep (which is open-web via Perplexity).
live-canvas
Conduct design interviews, generate UI variations, and collect live click-to-annotate feedback that streams into the session so edits land without leaving the browser. Use when the user wants rapid iterative UI refinement, not just batched feedback.
wikiweb
Command "wikiweb" from kimsanguine/llm-brain, covering 실행, 제공 기능 and 데이터가 없다면 (선택).
seo-compare
Compares the SEO of two pages or your page against a competitor. Shows differences in meta tags, structure, content, schema and speed side by side. TRIGGER when the user enters /seo-compare or asks to compare pages, analyze a competitor or see SEO differences between two URLs.
seo-crawl
Simulates how a crawler sees the site — reviews robots.txt, noindex directives, duplicate canonicals, redirect chains and accidentally blocked pages. TRIGGER when the user enters /seo-crawl or asks to review crawlability, indexing or how Google sees the site.
seo-structure
Reviews the site architecture — URL hierarchy, internal linking, click depth, orphan pages and internal authority distribution. TRIGGER when the user enters /seo-structure or asks to analyze the architecture, structure or internal linking of the site.