Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/hoangsonww/AI-News-Briefingnpx agentmods add skills/hoangsonww/ai-news-briefing/eval-driftWrote 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/hoangsonww/ai-news-briefing/eval-drift)<a href="https://agentmods.dev/skills/hoangsonww/ai-news-briefing/eval-drift"><img src="https://agentmods.dev/badge/skills/hoangsonww/ai-news-briefing/eval-drift/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/hoangsonww/ai-news-briefing/eval-drift"><img src="https://agentmods.dev/badge/skills/hoangsonww/ai-news-briefing/eval-drift.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00053 | $0.00436 |
| Opus 5 | $0.00026 | $0.00218 |
| Sonnet 5 | $0.00011 | $0.00087 |
| Haiku 4.5 | $0.00005 | $0.00044 |
Grade A, and why
eval-drift 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 12d 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.
What it actually says
Eval — Drift Detection
Surface quality slides before readers notice them. Robust to small-sample outliers via median + MAD scaling.
How to invoke
make eval-drift D=YYYY-MM-DD # status: ok / alert (informational)
make eval-drift D=YYYY-MM-DD ALERT_EXIT=1 # exit 3 on alert (cron-friendly)
Direct invocation supports tuning windows and thresholds:
python3 eval/drift.py --as-of YYYY-MM-DD \
--short-window 7 --long-window 30 \
--z-thresh 1.5 --streak 2 \
--exit-nonzero-on-alert
Algorithm
For each of the last --streak days (default 2):
short_med = median(last 7 days of composites)
long_med = median(last 30 days of composites)
long_mad = median(|x - long_med|)
scale = max(long_mad, 0.05) # floor avoids div-by-zero on flat history
z = (short_med - long_med) / scale
A day is "bad" when z < -1.5. If every day in the streak is bad, status flips to alert.
Output
JSON blob with status (ok / alert / no_data), the medians, the MAD, the z-score, and an alerts list of offending days.
What to tell the user
Lead with the status. If alert, show the offending dates and the z-scores; recommend running make eval-show and inspecting the per-card notes to identify which axis is dragging quality down. If ok, report the rolling medians anyway so the user has a sense of current quality.
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.
- 12d ago First seen · 46 lines · 53 tokens per session scan A a62160f706da
eval-drift is a skill published in the GitHub repository hoangsonww/AI-News-Briefing (42 stars, last pushed 4d ago), licensed MIT. It adds 53 tokens to every session and 436 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 2026-08-30.
Other skills, from other repositories
project-context
Detect a repository's languages, frameworks, databases, and observability stack, then load only the relevant Magician lore cores and task-matched deep dives. Use before implementation, debugging, review, performance, database, or logging work when project-specific technical guidance would improve the result, or when…
brainstorming
Interactive spec generation — turn ideas into concrete specs with R-numbered requirements and testable acceptance criteria.
executing
Autonomous task implementation workflow — implement, test, review, commit for each task in the frontier.
backpropagation
Trace runtime bugs back to spec gaps — identify missing acceptance criteria, update specs, generate regression tests, and detect patterns.
design-system
DESIGN.md integration for Forge — ensures visual consistency across all UI tasks through standardized design specifications.
planning
Decompose approved specs into ordered task frontiers with dependency DAGs, token estimates, and repo tags.