Use AIO Sandbox for isolated browser workflows.
Skills in this repository
Abd0r/porcupineai - Page 2
SkillsMP has collected 73 skills from Abd0r/porcupineai. Open a skill to review its source and details.
Abd0r/porcupineaiShowing 33 of 73 collected skills.
Safe cross-platform screen observation and UI interaction on macOS, Linux, and Windows.
Run a data-quality gate before drawing conclusions — inspect distributions, check for missing/duplicate/sensitive values, validate schema, and aggregate honestly.
Parse, validate, transform, extract from, and safely mutate JSON, CSV, and YAML using node -e and jq when available.
Reproduce bugs, isolate cause, fix, verify.
Reproduce, isolate, root-cause, and fix bugs with evidence instead of guessing. Use for any test failure, stack trace, regression, or unexpected behavior before proposing a fix.
Orient in an unfamiliar codebase — map the repo, locate symbol definitions vs usages, trace call paths, and understand build/test setup before editing.
Write and maintain README, CHANGELOG, guides, and code comments that stay accurate and current — verified against source, never imagined. Use when creating or editing any user-facing or developer-facing docs.
Safely and efficiently read, search, and edit files with Porcupine's filesystem tools, avoiding destructive mistakes.
Recover safely from broken Porcupine customizations.
Delegate focused, self-contained work to an isolated sub-agent with its own context, budget, and tools.
Decompose a large goal into targeted parallel sub-agent missions and integrate verified results — the harness carries the orchestration, the user states the goal.
The universal problem-solving loop for ANY task — understand, recon, decompose, plan, execute one step at a time, verify each, reflect. The umbrella that routes to domain skills. Use as the default operating loop when a task is not trivial.
Debug systematically instead of guess-and-patching — reproduce first, isolate one variable, read the actual error at file:line, and verify the fix with a real command. Use for any bug, test failure, or unexpected behavior.
Isolate risky work before it can hurt you — run untrusted code in a container, put experiments in a disposable clone, and use Porcupine's mode settings and Escape-cancel to contain yourself. Use whenever work could be destructive, untrusted, or hard to…
Write secure-by-default code — keep secrets out of code and logs, validate all input against an allow-list, and neutralize path traversal and command injection. Use when writing or reviewing any code that touches input, files, or the shell.
Compare methods fairly - baselines, held-out data, leakage prevention, and honest significance.
Analyze data without fooling yourself - hygiene, distributions, statistics, and honest plots.
Evidence-graded literature search, dedupe, and synthesis with citation trails.
Run experiments anyone can reproduce - seeds, pinned env, versioned data, full-run verification.
Write papers where every claim maps to evidence and every citation resolves.
Write reliable shell commands — correct quoting, pipes, exit-code checks, and knowing when to use a purpose-built tool instead of bash.
LOC and language breakdown with pygount or cloc.
Safe local git plus pointer to GitHub skill stack.
Set up GitHub auth via gh CLI, HTTPS token, or SSH.
Review PR diffs and leave gh review comments.
Explore a GitHub repo structure, README, and layout.
List, create, triage, label, close GitHub issues with gh.
Resolve PR merge conflicts in isolated worktrees.
PR lifecycle: branch, commit, push, open, CI, merge.
Profile a GitHub user from public repos and README.
Clone, fork, create repos; remotes, releases, settings.
Verify which issues a PR solves via gh, not web search.