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 instructions/ourobx/agent-ebpf/self-learninggit clone --depth 1 https://github.com/ourobx/agent-ebpfWhat 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 | $0.02381 | $0.02381 |
| Opus 5 | $0.01190 | $0.01190 |
| Sonnet 5 | $0.00476 | $0.00476 |
| Haiku 4.5 | $0.00238 | $0.00238 |
Grade C, and why
agent-ebpf self-learning.instructions.md scanned grade C with 1 finding 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.
Hidden instructionshighPrompt injection
Directives inside HTML comments, invisible characters or bidirectional overrides are read by the model and not by the person reviewing the file.
<!-- Copilot Instruction Telemetry: This file was auto-generated by Claude Skills Manager --> How it starts
The opening of the file, as written. The whole thing — 212 lines — stays where its author put it; the contents beside it link to each section on GitHub.
self-learning
Self-Learning
A small, project-agnostic accumulated-experience store. The goal: the second time something fails, fixing it is instant because the fix is already written down — and reliable commands don't need re-verifying every session.
Storage layout
All state lives under .claude/learning/ in the target project (create it on
first use):
.claude/learning/
runs.jsonl append-only log of recorded outcomes (gitignore this)
patterns.md auto-generated report (gitignore this)
session-learnings.md human/agent-curated decisions and fixes (commit this)
knowledge-cache.md cached answers to repeated questions (commit this)
skill-feedback.jsonl user negative reactions to agent/skill behavior (gitignore)
task-skill-proposals.json proposed skills for the current task (gitignore)
Add .claude/learning/runs.jsonl, .claude/learning/patterns.md,
.claude/learning/skill-feedback.jsonl, and .claude/learning/task-skill-proposals.json to
.gitignore if not already ignored — they're machine-local history.
session-learnings.md and knowledge-cache.md should be committed: they're
durable, reviewable output.
Run record schema (one JSON object per line in runs.jsonl)
{"ts": "2026-06-11T14:32:00", "skill": "terraform-plan-review", "action": "plan",
"rc": 0, "duration": 4.2, "error": "", "hint": "", "note": "", "tokens": 12345,
"metadata": {"invoked": true}}
skill/action: a short identifier for what was run (e.g. skill name + subcommand, or"task"+ a short task name).metadata.invoked: set totruewhen this skill was actually invoked in the session (not merely listed in context). Cost attribution uses this to distinguish active skills from enabled-but-unused skills.rc: 0 for success, non-zero for failure.error: first meaningful error line (truncate to ~200 chars), empty on success.hint: a short fix description if one is known (see "Deriving hints" below); empty if none.note: optional free-text context.tokens: optional total token count (input + output + cache write + cache read) attributable to this run — see "Recording token usage" below. Omit if it can't be determined.
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 · 212 lines · 2,381 tokens per session scan C 5e6981aee098
agent-ebpf self-learning.instructions.md is an instructions file published in the GitHub repository ourobx/agent-ebpf (1 stars, last pushed 7d ago), licensed MIT. It adds 2,381 tokens to every session, about $0.0119 per session on Opus 5. A static security scan graded it C with 1 finding (hidden instructions). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
casbin-gateway CLAUDE.md
Claude Code instructions for apache/casbin-gateway, covering claude.md and code style.
aigis CLAUDE.md
Instructions for killertcell428/aigis, covering aigis — project guidelines, zenn記事管理（zenn cli + github連携）, ディレクトリ構成, 記事の作成 and プレビュー.
preloop AGENTS.md
AGENTS.md instructions for preloop/preloop, covering preloop development guide, commands, git workflow, keep sample data generic and code style.
pfsense-mcp-server AGENTS.md
Instructions for night4me/pfsense-mcp-server, covering pfsense mcp server, architecture and security, development workflow, long-running validation and test parallelism.
xops-cli AGENTS.md
Instructions for wentf9/xops-cli, covering ai 助手开发协作指南 (agents.md), 🎨 1. 编码规范 (coding standards), 📦 2. git 规范 (git standards) and 🛠️ 3. 测试、构建与 ci (testing, building & ci).
fwskillsshare AGENTS.md
Instructions for fastrevmd-lab/fwskillsshare, covering firewall skills repository instructions, purpose and architecture, setup and development, required checks and skill description constraints.