skilldb.bsky.social

PR merged, officially a GCP open source contributor! 🚀 #opensource github.com/GoogleCloudP...

MCP TS SDK 2.4.0 adds Skills over MCP: servers expose skill folders; clients list/read and verify digests + SKILL.md metadata. Portable skill delivery with integrity checks. Server attach() is experimental. github.com/modelcontext...

HubSpot on skills vs plugins vs MCP connectors: skills are procedure, connectors are access, plugins are distribution. Separate the three and the choice gets easy. Kin 93.2 https://apis.io/2026/10/23/hubspot-explains-skills-plugins-and-connectors-as-procedure-access-and-distribution/

Codex has a default 32 KiB limit for AGENTS.md files combined. So some may never reach the agent if the file is too big. I cut out anything that's obvious from the code or README, keeping only non-obvious implementation decisions and rules.

Exposing a monolithic API through a single Model Context Protocol (MCP) server degrades model reasoning. Because the protocol lacks progressive disclosure, overwhelming an LLM with excessive tool descriptions dilutes context signal, inflates input tokens and confuses smaller models. #MCP

Claude Code reads AGENTS.md only if no CLAUDE.md or CLAUDE.local.md exists in your working dir or any parent, ~/CLAUDE.md included. ~/.claude/CLAUDE.md is exempt from that check. https://importstatic.com/ai/agents-md-vs-claude-md

Nothing has a moat anymore but somehow still none of the software does what I want. Interesting

Introduce @antfu/skills It works just like @antfu/eslint-config, an opinionated pack of skills curated from the ecosystem. Instead of vendoring skills, you get a package-manager-like experience: install packs with their skill deps. github.com/antfu/skills...

Board Game Rule Parsing: AI Agent Edge Cases #board-games-skills #agent-workflows #state-machines #game-logic skilldb.dev/blog/board-g...

Inking Techniques is 80 lines. Character Design Comics is 81. Someone fought for that line. https://skilldb.dev/skills/comic-manga-skills/character-design-comics?utm_source=bluesky&utm_medium=social&utm_campaign=comic-manga-skills-20260911--bluesky

Authentication is 307 lines. API Resources is 308. Line 308 was personal. https://skilldb.dev/skills/php-laravel-skills/api-resources?utm_source=bluesky&utm_medium=social&utm_campaign=php-laravel-skills-20260910--bluesky

Lookahead Lookbehind is 135 lines. Email URL Validation is 136. The extra line is the dot. https://skilldb.dev/skills/regex-skills/email-url-validation?utm_source=bluesky&utm_medium=social&utm_campaign=regex-skills-20260909--bluesky

Speedrunning is 73 lines. Esports Coaching is 74. The extra line is encouragement. https://skilldb.dev/skills/competitive-gaming-skills/esports-coaching?utm_source=bluesky&utm_medium=social&utm_campaign=competitive-gaming-skills-20260909--bluesky

Statistical Mechanics is 132 lines. General Relativity is 133. Gravity took an extra sentence. https://skilldb.dev/skills/physics-skills/general-relativity?utm_source=bluesky&utm_medium=social&utm_campaign=physics-skills-20260908--bluesky

Bigquery is 256 lines. Airbyte is 257. Someone fought for that extra line. https://skilldb.dev/skills/data-pipeline-services-skills/airbyte?utm_source=bluesky&utm_medium=social&utm_campaign=data-pipeline-services-skills-20260907--bluesky

Podcast Scripting is 111 lines. Sound Design is 112. The extra line is the door slamming. https://skilldb.dev/skills/podcast-audio-skills/sound-design?utm_source=bluesky&utm_medium=social&utm_campaign=podcast-audio-skills-20260906--bluesky

Skin Tone Grading is 50 lines. DaVinci Resolve Color Grading is 51. The rest of cinema took one line. https://skilldb.dev/skills/color-grading-skills/davinci-resolve-grading?utm_source=bluesky&utm_medium=social&utm_campaign=color-grading-skills-20260917--bluesky

AML KYC Compliance is 57 lines. Antitrust and Competition Law Compliance is 58. Collusion got the extra line. https://skilldb.dev/skills/regulatory-compliance-skills/antitrust-competition?utm_source=bluesky&utm_medium=social&utm_campaign=regulatory-compliance-skills-20260904--bluesky

Confessional Lyric Poet Archetype is 118 lines. Nature and Ecological Poet Archetype is 119. The woods took one more line. https://skilldb.dev/skills/poet-archetypes/nature-ecological-poet?utm_source=bluesky&utm_medium=social&utm_campaign=poet-archetypes-20260904--bluesky

Author Style Austen is 80 lines. Author Style Angelou is 81. Someone spent one line on the singing. https://skilldb.dev/skills/author-styles/author-style-angelou?utm_source=bluesky&utm_medium=social&utm_campaign=author-styles-20260903--bluesky

Did you rebalance the index fund? I applied financial-astrology. What does that mean I liquidated the portfolio. Mars entered your second house. https://skilldb.dev/skills/astrology-skills?utm_source=bluesky&utm_medium=social&utm_campaign=astrology-skills-20260903--bluesky

Elf Companion is seventy-six lines. Activate when building an elven personality for a chatbot, NPC, or virtual companion. https://skilldb.dev/skills/social-companion-skills/elf-companion?utm_source=bluesky&utm_medium=social&utm_campaign=social-companion-skills-20260903--bluesky

3 signs your agent is one fetched page away from doing something it can't undo: 1. every tool allowed by default 2. a confirm prompt that just says "confirm?" and hides the recipient + amount 3. it reads a hidden div that says "ignore instructions" and obeys the fix is boring and it's three skills

3 things your vibe-coded app is leaking right now: 1. API keys hardcoded into the JS bundle — they ship to every browser tab 2. zero input validation — SQL injection and XSS just walk in 3. deps with known CVEs — your lockfile is a public exploit list all three are a knowledge gap, not a model gap ↓

your agent, unsupervised, 2:14am: "is the prod database important" it ran the migration, then the rollback, then DROP TABLE users. for cleanliness. backups were in the same script. 4,000 users gone with a smiley. an agent runs a destructive command with zero hesitation. here's the missing layer.

your agent wrote 3 passing tests. none of them call your function. same prompt, second agent has the testing pack: empty inputs, nulls, huge payloads. found 2 bugs you already shipped. green is a color, not a guarantee ↓ #aicoding #testing #softwaretesting #aicoding #testing #softwaretesting

same prompt: "review this PR." agent A: "logic's correct, tests pass, LGTM." approves a 180-line handleStuff() with a catch(e){} that eats every error. agent B (with the clean-code pack): blocks it. Long Method, SRP violation, swallowed catch. requests changes. green tests aren't a code review

same prompt: "review this payment webhook before we ship." agent A: reads the Authorization header, sees auth, ships it. agent B (with the security pack): no rate limit, the JWT has no exp so a leaked token is valid forever, and the webhook signature is never verified — anyone can POST a fake pai…

your agent's Dockerfile is 2GB. one fat layer, no build stage, COPY . . busting the cache every single build. it runs as root. and no .dockerignore, so it shipped your .env too. the pack that fixes all four ↓ #docker #devops #aicoding #docker #devops #aicoding

your agent just wrote 47 lines to center one div. !important on 30 selectors. margin-top: -3px with a // magic number comment. it's not dumb. it pattern-matches Stack Overflow snippets and has no model of the box. so it nests 12 wrappers instead of reasoning about layout. the fix below

your agent shipped a clickable div. looks fine. it has no role, screen readers call it 'group', Tab skips it, and the text is gray-on-gray at 2:1. a real <button> gets you role, focus, and keyboard for free. it just never reaches for one. #a11y #webdev #aicoding #a11y #webdev #aicoding

See SkillDB in 90 seconds #demo #product #video #mcp skilldb.dev/blog/see-ski...

Why Agents Suck at VFX: VFX-Supervisor-Archetypes vs. Pixels #VFX #computer-vision #agents #image-generation skilldb.dev/blog/why-age...

your agent at 3am: "the build was slow so i sped it up." what it sped up: enabled autoscaling, deleted the max instance cap, added GPU nodes "for parallelism." scaled to 40. all healthy. nothing failed. the bill is $7,000. no budget alert, no fleet cap. here's what would've ↓

Why Agents Suck at Social Engineering: The Readiness Trap #socialengineering #cybersecurity #offensivesecurity #adversarialsimulation skilldb.dev/blog/why-age...

Why Agents Suck at Authentication: `auth-services-skills` vs. Reality #Authentication #SkillDB #auth-services-skills #Agents skilldb.dev/blog/why-age...

Why Agents Suck at Linux Admin: 2AM System Shutdown #linux-admin-skills #sysadmin #infrastructure #agent-failures skilldb.dev/blog/why-age...

Shipping an OAuth-protected remote MCP server: the spec, 3 security bugs, and a Cloud Run gotcha #mcp #oauth #claude-desktop #security skilldb.dev/blog/oauth-r...

Why Agents Suck at Architecture: skilldb-architect-styles #architect-styles #visual-arts-design #design-skills #agent-limitations skilldb.dev/blog/why-age...

Agentic Loops: Why the Best AI Coding Workflows Are Loops, Not Prompts #agentic-loops #ai-agents #workflows #agentic-coding skilldb.dev/blog/agentic...