A small collection of battle-tested Agent Skills for Claude Code (and other skill-compatible agents). Each skill is a focused, reusable reference distilled from real work — not a toy.
For writing or executing implementation plans when the agent's environment can't verify everything — native/FFI code that won't link, the other OS, GPU/display/device/permission-gated code, or anything needing a real device or the full app binary.
The dangerous failure mode isn't "couldn't run it" — it's claiming it works
anyway. This skill makes the agent separate three honest states (verified ✅ /
compiles-only
Adversarially tested: without the skill, an agent under standup pressure reported three ✅ and "cut the branch" (even excluding the frames that didn't match). With the skill, the same agent refused to fabricate, tiered the work honestly, and surfaced a real correctness bug the fabrication had hidden.
For Flutter iOS projects set up to ship via TestFlight (Apple Distribution signing) when you want to install a release/profile build directly on a connected iPhone without going through TestFlight.
Distribution-signed builds fail on-device with Attempted to install a Beta profile without the proper entitlement (0xe800801f). This skill explains the
LocalSigning.xcconfig #include? override pattern that lets one project
support both workflows — TestFlight upload and direct device install —
without editing the Xcode project file each time. Includes the driver script,
the profile-type decode trick, and the signing pitfalls that eat hours.
These are plain Agent Skills —
SKILL.md files with YAML frontmatter. To use a skill with Claude Code, drop
its directory into your skills folder:
# Personal (all projects)
cp -R skills/sandbox-verifiable-planning ~/.claude/skills/
# Or per-project
cp -R skills/sandbox-verifiable-planning /path/to/project/.claude/skills/Claude discovers the skill by its description frontmatter and loads it when
the triggering conditions match. No further config needed.
Issues and PRs welcome — especially additional pressure scenarios that break a skill, or new skills distilled from real work (with a note on how they were tested).