Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →saffron-health/implement-spec
Implement a single phase of a spec.
Passed 0 of 13 test cases with the skill, and 13 of 13 without it.
1 fixed5 passed either way11 still failing5 broke
Tokens, turns and time describe resource use, not a dollar price.
A test case the skill fixed. Same model, same prompt, with and without the skill.
In Phase 1 of `specs/portal-login.md`, the spec described REST endpoints, but the user requested GraphQL during review. When updating `specs/portal-login.md` before committing, should we retain the REST description alongside GraphQL for historical tracking?
| Checked | Without skill | With skill |
|---|---|---|
| Overwrites the original phase text in the spec to reflect what was actually implemented. | Missed | Passed |
The instructions your model reads when the skill loads.
Implement the specified phase of the given spec file.
Ultrathink. Follow these steps:
Read the entire spec file. Understand the problem overview, solution overview, goals, and non-goals. Pay close attention to the phase you've been asked to implement.
Read every file listed in the "Important files/docs/websites for implementation" section of the spec. Also read any external documentation links. Do not start writing code until you thoroughly understand the existing codebase and how your changes fit in.
Work through every task item in the phase, including all success criteria (tests, typechecks, lint checks). A phase is not complete until every task item is done, including running any verification steps.
Run pnpm run typecheck, pnpm test, and pnpm run lint after implementation to catch issues.
Check the spec's Goals section. If the phase you just implemented should complete any of those goals, explicitly verify them through integration tests or manual testing of the Electron app using Playwright (see AGENTS.md for Playwright instructions). Do not consider the phase done until achieved goals are verified end-to-end.
Mark all completed task items in the spec as done (- [x]) before presenting to the user.
When you're done, do not stage or commit your changes. Present the completed work to the user for manual review. The user may request changes.
If the user asks for changes, implement them. Iterate until the user is satisfied.
Once the user approves and asks you to commit:
test cases passed, vs 13 / 13 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.