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 skills/netgrade-digital/shared-agents/capture-learningnpx skills add netgrade-digital/shared-agents --skill capture-learninggit clone --depth 1 https://github.com/netgrade-digital/shared-agentsWhat 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 | $0.00050 | $0.01161 |
| Opus 5 | $0.00025 | $0.00580 |
| Sonnet 5 | $0.00010 | $0.00232 |
| Haiku 4.5 | $0.00005 | $0.00116 |
Grade A, and why
capture-learning scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 133 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Capture Learning
Write team learnings to pending/ — drafts awaiting human review.
Ask first (required)
After non-trivial tasks, the agent must ask:
„Soll ich ein Team-Learning in shared-agents anlegen?"
Only write if the user says yes. Never silently write to pending.
When content is worth capturing
Propose when all apply:
- Insight is reusable beyond this session
- Would help a teammate or future agent
- No secrets or customer-specific data
Skip when trivial, already documented, or user said no.
Where to write (pending only)
Canonical path only — see docs/canonical-paths.md.
# Resolve before write (never use workspace-relative paths):
sa pending path YYYY-MM-DD-short-slug
# Or: "$SHARED_AGENTS_HOME/scripts/learning-path.sh" YYYY-MM-DD-short-slug
Mandatory
- Write only under the path from
sa pending path(typically$SHARED_AGENTS_HOME/team/learnings/pending/). - Use the absolute path in the Write/edit tool — not paths from the open project or Core dev checkout.
- Do not commit learnings into the public Core-Repo.
Forbidden
Development/Work/shared-agents/learnings/(Core has no learnings)- Customer project repos (
.cursor/, projectdocs/, etc.) approved/(human/PR only)
Never write to approved/ — that is human/PR territory.
File format
---
id: project-YYYY-MM-short-slug
project: project-name
domain: [tag1, tag2]
tags: [keyword1, keyword2]
versions: [shopware:6.6.10, php:8.3.14]
confidence: high
source: task
created: YYYY-MM-DD
author: github-or-name
---
## Kontext
What problem or situation triggered this.
## Erkenntnis
The reusable insight in 1–3 sentences.
## Anwendung
Concrete steps or rule of thumb for next time.
## Links
- path/to/file or PR URL (optional)
Versions (required when a stack applies)
Always set versions: for framework/runtime-specific learnings. Use [] only when no stack version is relevant (e.g. pure process/infra docs).
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.
- 2d ago First seen · 133 lines · 50 tokens per session scan A 4fc75af09379
capture-learning is a skill published in the GitHub repository netgrade-digital/shared-agents (4 stars, last pushed 2mo ago), licensed MIT. It adds 50 tokens to every session and 1,161 once invoked, about $0.0003 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 2026-08-31.
Other skills, from other repositories
memory-curation
When you have read / processed a workspace asset in this session and learned something durable about it, write a memory page so future sessions benefit. Maintain the workspace wiki's hierarchical structure as it grows.
speak-human-tw
「說人話」：繁體中文的去 AI 味改寫 skill。審查與改寫文字，去除 AI 味道、校正中國用語與半形標點，讓文字讀起來像真人寫的。 觸發時機：用戶說「去 AI 味」「說人話」「這段好 AI」「改自然一點」「幫我潤稿去掉 AI 感」「校對一下再發」，或要求檢查電子報、社群貼文、銷售頁、課程文案、客服回信、簡報、公告、Email 等對外文字的語感。 不要觸發：逐字翻譯、模仿特定品牌 voice、事實查核（非風格問題）、程式碼／log／設定檔、要求「潤成雷蒙的語氣」（那是 content-writing skill 的事，本 skill 只去 AI 味、不加個人風格）。.
docs-workflows
Documentation and extension workflows derived from local slash commands (docs, explanation, issues, prototypes, tutorials, MCP design review).
init-workspace-documentation
Skill "init-workspace-documentation" from griddynamics/rosetta, covering agent memory.md, agent memory, preventive rules, what worked and what failed.
editorial-illustrations
Generate meaning-carrying editorial data-illustrations in the monotykamary / Linear aesthetic (near-black grayscale, Inter display + mono labels, hairline framed figures) with a single coral accent. This is a GENERATIVE GUIDE, not a template gallery: it teaches the "claim -> geometry" method so any session can invent…
Academic CV Builder
Format CVs for academic positions with publications, grants, and teaching.