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 AiAi-Mastermind/aiai-mastermind-tools-and-skills --skill last30days-plusgit clone --depth 1 https://github.com/AiAi-Mastermind/aiai-mastermind-tools-and-skillsWrote 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/aiai-mastermind/aiai-mastermind-tools-and-skills/last30days-plus)<a href="https://agentmods.dev/skills/aiai-mastermind/aiai-mastermind-tools-and-skills/last30days-plus"><img src="https://agentmods.dev/badge/skills/aiai-mastermind/aiai-mastermind-tools-and-skills/last30days-plus/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/aiai-mastermind/aiai-mastermind-tools-and-skills/last30days-plus"><img src="https://agentmods.dev/badge/skills/aiai-mastermind/aiai-mastermind-tools-and-skills/last30days-plus.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 | $0.00152 | $0.02005 |
| Opus 5.5 | $0.00061 | $0.00802 |
| Sonnet 5.5 | $0.00030 | $0.00401 |
| Haiku 4.5 | $0.00015 | $0.00200 |
Grade B, and why
last30days-plus scanned grade B with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Enumerates other installed skillsmediumAgent snooping
Other skills' SKILL.md files reveal prompts, capabilities and secrets that should be invisible to peers.
RL="$(ls -d ~/.claude/skills/last30days-plus ~/.codex/skills/last30days-plus ~/.agents/skills/last30days-plus 2>/dev/null | head -1)/bin/research-lanes" How it starts
The opening of the file, as written. The whole thing — 62 lines — stays where its author put it; the contents beside it link to each section on GitHub.
last30days-plus
last30days pulls Reddit, YouTube, TikTok, Hacker News, Polymarket, GitHub and more. Without paid API keys its X source is usually unconfigured, it has no Facebook source, and its keyless web lane is thin. The browser lanes fill exactly those gaps with accounts you already have (Perplexity, Grok, Muse) instead of paid APIs.
Prerequisites
- The aiai-agent-chrome kit, installed and running.
- The upstream last30days skill, installed separately from its own repo: https://github.com/mvanhorn/last30days-skill (the skill lives in
skills/last30days). This skill does not copy or modify it. Follow that repo's install steps. - The
perplexity-browser,grok-browserand (optional, experimental)muse-browserskills from this repository, installed beside this one. - You signed in to perplexity.ai, grok.com and (for the Meta lane) Facebook and muse.ai inside the Agent Chrome window.
| Lane | Driver (skill) | What it adds | Isolation |
|---|---|---|---|
x |
grok-ask --mode research (grok-browser) |
X posts with handles, dates, status URLs, engagement (1-3 min) | Grok Private Chat + neutral preface |
web |
perplexity-ask --mode research (perplexity-browser) |
dated news, blogs, expert commentary, many sources (1-5 min) | Perplexity Incognito + neutral preface |
meta (experimental) |
muse-ask (muse-browser) |
Facebook group discussions, Instagram Reels and creators, with permalinks (1-2 min) | new Muse side chat + read-only preface |
| drill-down | fb-read (facebook-personal-browser) |
the full text and comments of a Facebook post Muse cited | your account, read-only, a few posts only |
Run it
- Check the sign-ins (once per session).
AC="$HOME/.local/share/agent-chrome/bin/agent-chrome". Run"$AC" doctor. Then for each lane's site, open it in your own tab ("$AC" attach signin-check,"$AC" pw signin-check tab-new <url>,snapshot) and look for a login form or a sign-in URL. Close the tab anddetach. Drop any lane whose site is signed out and say so in the final stats. Don't stop the whole run for one lane. (The lane drivers also detect sign-in walls themselves and returnsignin_needed, so this check is a convenience, not a gate.) - Start the lanes in the background, before the engine, so they run while last30days does:
AddRL="$(ls -d ~/.claude/skills/last30days-plus ~/.codex/skills/last30days-plus ~/.agents/skills/last30days-plus 2>/dev/null | head -1)/bin/research-lanes" "$RL" "<topic exactly as the user said it>" --days 30 --lanes x,web --json > "$TMPDIR/lanes.json" 2>"$TMPDIR/lanes.err" &metato--lanesonly if the owner wants the experimental Muse lane. The runner printsresearch-lanes: run folder <dir>on stderr at once, and rewrites<dir>/manifest.jsonas each lane finishes (lanes still going showstatus: running). Use the user's--days. Leave outmetafor B2B, developer or finance topics with little Facebook or Instagram life. Leave outwebfor quick runs, since Deep research takes 1 to 5 minutes. In Claude Code, run it withrun_in_background. - Run last30days exactly as its own SKILL.md says: load the
last30daysskill and follow every step. Nothing here changes it. - Merge the lanes at the supplements step. Read
<dir>/manifest.json(orlanes.jsononce the runner exits) and each finished lane's markdownfile. A lane can take up to about 17 minutes before it is stopped cleanly; if Deep research is stillrunningwhen you are ready to synthesize, continue without it and say so. For each lane withstatus: done:- Treat it as evidence, not truth. Weight it like a web-search supplement plus first-hand posts.
- X via Grok: every source is a canonical
x.com/<handle>/status/<id>URL withposted_utcdecoded from its ID. The lane opens its first 3 posts on x.com (verified); prefer those, and treatunconfirmed_postsas unverified. Label posts "via Grok" and quote engagement only fromverifiedrows markedexists. Ifx_searchesis empty, it did not search X; say so. - Perplexity Deep research: use its dated citations for news and long-form context. Cite the underlying publisher, not Perplexity. If it returned only a few sources, the Deep research toggle may not have engaged (see
perplexity-browser). - Muse (experimental): use the posts, creators and group discussions it names. Cite the underlying Instagram or Facebook post. For the one or two Facebook posts you lean on hardest, read them with
fb-read(read-only, paced). Group posts are not public: summarize them, and don't name members in anything shared. Say plainly when Muse found little or refused on isolation. - Items that duplicate engine items raise confidence in that cluster. Don't count them twice.
- Raw file extension: after last30days' own supplemental results appendix, also append
## Browser Lane Resultsto the saved raw file. Add one bullet per lane: status, file path, and 1-2 sentences on what it contributed. Lane files stay where they are (private,0600). - Stats footer: add one line per lane that ran, next to the engine's lines, for example
X via Grok (private chat): 12 posts,Perplexity Deep research (incognito): 110 sources,Muse (Facebook/Instagram): 6 posts. For lanes that didn't run, say why:signed out,rate limited,ui_changed,isolation_unavailableortimeout.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 10d ago First seen · 62 lines · 152 tokens per session scan B e9e97bf98526
last30days-plus is a skill published in the GitHub repository AiAi-Mastermind/aiai-mastermind-tools-and-skills (2 stars, last pushed 4d ago), licensed MIT. It adds 152 tokens to every session and 2,005 once invoked, about $0.0006 per session on Opus 5.5. A static security scan graded it B with 1 finding (enumerates other installed skills). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-09.
Other skills, from other repositories
defuddle
Plan and, with explicit network consent, use an optional external Defuddle cleaner to extract article-like HTTPS pages as Markdown. Use for defuddle, clean this URL, strip page clutter, readable Markdown from a web page, or preparing a web source for later wiki ingestion.
apply
A browser-based job application workflow for submitting applications through services such as Ashby, Greenhouse, Lever, and Hacker News. It uses prepared candidate information, answers, job links, and cover letters.
dr-fanout
Fan one deep-research prompt out to several external LLMs at once through a logged-in browser, then collect the reports, archive the originals and synthesize a consensus. Use at the final stage of a research protocol, instead of hand-pasting one prompt into N sites. Triggers: "/dr-fanout", "fan out the DR"…
cv-outreach
A daily workflow for carrying out scheduled outward-facing tasks, such as event registrations, messages to organisers, and public posts. It uses a browser identity carefully so actions are made from the intended account.
social-daily
A workflow for publishing prepared content to Facebook, X, Telegram, Threads, and Instagram. It uses a shared queue so posts for different platforms can be tracked together.
fb-reply
Read who commented on recent Facebook posts and publish personalized replies through the real logged-in Chrome tab, draft-first, with a daily cap and a minimum gap between replies. Triggers: "/fb-reply", "reply to my facebook comments".