Install
$ agentstack add skill-int2t05-engineering-skills-implement ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Implement
Execute the work described by a spec or tickets in thin vertical slices. Each slice cuts a complete path through every layer (schema, API, UI, tests) and lands green. Drive TDD within each slice; verify framework-specific patterns against official docs; subject non-trivial decisions to a fresh-context doubt cycle before they stand.
When to use
- Implementing work described by a spec, plan, or set of tickets.
- Multi-file changes that should land one vertical slice at a time.
- Framework-specific code where current docs determine correct patterns.
- High-stakes or unfamiliar code where correctness matters more than speed.
- Lightweight changes that don't need a spec — small config edits, mechanical renames/moves, project scaffolding (see
references/lightweight-changes.md).
Not for: changes needing design decisions (use spec); system architecture (use architecture); pure test-driven exploration or bugfix-via-test with no spec — red-green-refactor as the whole task (use tdd). Have a spec and want slice-by-slice implementation? That stays here, with TDD driven within each slice.
Steps
- Read the spec or tickets. Orient on what "done" means. Note any
pre-agreed TDD seams — integration points where tests must exist before code lands. If versions are missing or ambiguous, ask; the version determines which patterns are correct.
- Slice the work vertically. One demoable end-to-end path per slice, sized
to a single fresh context. Keep it compilable between slices. Wide refactors (one mechanical change fanning across the codebase) are the exception — sequence them as expand–contract, never force them into a tracer bullet.
- For each slice, implement via TDD at the seams:
- Write the failing test first (red), then the code (green), then refactor.
- For framework-specific code: detect the stack and exact versions from the
dependency file; fetch the relevant official docs page (not the homepage); follow the documented pattern and cite the source URL in a code comment. Flag anything you cannot verify as UNVERIFIED. When docs conflict with existing code, surface the conflict — don't silently pick one. See [references/source-verification.md](references/source-verification.md).
- Run the typecheck and the affected test file(s) after each slice. Don't
repeat a command on unchanged code — it adds no information.
- Run a doubt cycle for non-trivial decisions. A decision is non-trivial
when it introduces branching logic, crosses a module boundary, asserts a property the type system can't verify (thread safety, idempotence, ordering), or has irreversible blast radius (production deploy, data migration, public API change). Name the claim, extract the artifact + contract (not your reasoning), spawn a fresh-context adversarial reviewer, reconcile each finding against the artifact text, and stop on trivial findings / 3 cycles / user override. See [references/doubt-cycle.md](references/doubt-cycle.md).
- Commit the slice with a descriptive message. Each slice is independently
revertable. Don't mix concerns — feature, refactor, and config changes go in separate commits (see §9, behavior-preserving change discipline, in engineering-principles.md).
- After all slices: run the full test suite once. Not per slice — once, at
the end. If it fails, localize with the debugging skill; don't re-run blind.
- Run
/code-reviewon the diff. Address blocking findings before
committing final state.
- Commit final state to the current branch (if not already committed per
slice).
Verify
- Full test suite passes.
- Typecheck clean.
/code-reviewreturned no blocking findings.- Framework-specific patterns carry source citations (or explicit
UNVERIFIED
flags); no deprecated APIs used.
- Non-trivial decisions survived at least one fresh-context doubt cycle.
- Work committed to the current branch.
References
- [${CLAUDEPLUGINROOT}/references/engineering-principles.md](${CLAUDEPLUGINROOT}/references/engineering-principles.md) — discipline shared by every skill.
- [${CLAUDEPLUGINROOT}/references/clean-code.md](${CLAUDEPLUGINROOT}/references/clean-code.md) — code-quality bar for the implementation.
- [references/source-verification.md](references/source-verification.md) — source hierarchy, citation rules, retrieval safety.
- [references/doubt-cycle.md](references/doubt-cycle.md) — the five-step doubt cycle and cross-model escalation.
- [references/lightweight-changes.md](references/lightweight-changes.md) — small edits, renames, scaffolding that don't need a spec
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: int2t05
- Source: int2t05/engineering-skills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.