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 skills add AbdurRafay2004/handoff --skill learngit clone --depth 1 https://github.com/AbdurRafay2004/handoffWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/abdurrafay2004/handoff/learn)<a href="https://agentmods.dev/skills/abdurrafay2004/handoff/learn"><img src="https://agentmods.dev/badge/skills/abdurrafay2004/handoff/learn/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/abdurrafay2004/handoff/learn"><img src="https://agentmods.dev/badge/skills/abdurrafay2004/handoff/learn.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What 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 5.1 | $0.00066 | $0.01746 |
| Opus 5.5 | $0.00026 | $0.00698 |
| Sonnet 5.5 | $0.00013 | $0.00349 |
| Haiku 4.5 | $0.00007 | $0.00175 |
Grade A, and why
learn 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 1mo ago.
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 — 148 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Learn
Part of the Learn phase — see the workflow skill for tiers and sequencing.
Work that ends without a recorded lesson gets re-learned the hard way. This skill closes the loop: reconstruct what actually happened (from git, not memory), find the cause, record what it taught us, and hand the user — a PM building development judgment — one thing they can carry forward.
Tier scaling — most work does NOT get a retro
| Situation | What Learn means |
|---|---|
| T1 / T2 | One CHANGELOG.md line + a STATUS.md update. Done. Do not run a retro; that would be ceremony the tier doesn't warrant. Only add a LEARNINGS.md entry if something genuinely surprising surfaced. |
| T3 completed | Full retro (below), even when it went well — lessons from success calibrate what to repeat. |
| Incident / rollback / user asks for a retro | Full retro, regardless of tier. |
The retro method
0. Anti-hallucination guard — verify before analyzing
The retro is built from git history, never from what you remember doing. Before analyzing anything:
git log --since="<window>" --format="%h|%ai|%s" --stat
- Confirm the commits you're about to discuss actually exist in that output. Never analyze a commit you cannot see the hash of.
- If the window returns zero or suspiciously few commits, stop and ask the user to confirm the date range or the branch — do not fabricate a coherent-sounding narrative from nothing.
- If the retro is about a specific incident, locate the exact commits
involved (
git log --follow <file>,git show <hash>) and quote them.
1. Blameless framing
Analyze systems, not people. Assume every decision was reasonable given what was known at the time; ask "why was that decision reasonable?" not "who messed up?" — the fix that matters is the one that makes the next attempt succeed. Blameless does not mean toothless: action items still get owners and dates, and hard truths still get said.
2. Timeline → root cause → contributing factors → action items
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.
- 1mo ago First seen · 148 lines · 66 tokens per session scan A 1437dc084a7e
learn is a skill published in the GitHub repository AbdurRafay2004/handoff (4 stars, last pushed 10d ago), licensed MIT. It adds 66 tokens to every session and 1,746 once invoked, about $0.0003 per session on Opus 5.5. 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 2026-08-31.
Other skills, from other repositories
training-plan
Design a legal compliance training program — role-based needs, modules with objectives, cadence, assessments, completion tracking and regulation-specific courses. Use when the user needs to train employees on privacy, contracts, anti-bribery or other legal duties and track completion. Not for security awareness…
legal-glossary
Generate a plain-language glossary of legal terms by area — contracts, privacy, IP, employment, compliance and corporate — with an example and why each term matters to non-lawyers. Use when the user or their team keeps hitting legal jargon and wants the terms explained in everyday words.
diffscope
A skill that turns recent code changes into a plain-language learning briefing. It shows what changed, how behavior differs, and explains relevant Python syntax.
python-syntax-tutor
A Python syntax tutor that explains unfamiliar language features in the code where they appear. Python is a programming language; examples include decorators, async code, generators, and type hints.
checkpoint
Write the current working state (task, decisions, files touched, open problems, next steps) to .amber/STATE.md so it survives compaction and session restarts. Use when the user asks to checkpoint/save state, before risky long operations, when context feels heavy, or before ending a work session.
focus
Declare a focus contract for the current task to keep context lean — scopes which files/dirs are in play, bans out-of-scope exploration, and routes broad searches to subagents. Use when starting a task, when the user says "focus on X", or when the session is drifting across unrelated parts of the codebase.