What I do
- •Turn requested workflows into reusable scripts under
scripts/. - •Normalize script implementation to project conventions (
*.sh.ts+ shebang + helpers). - •Capture successful ad-hoc command sequences as reusable script entrypoints.
Accepted inputs
Use this skill when script requirements come from any of these sources:
- •User prompt only (goal and constraints)
- •Provided example script
- •Provided bash script or shell command sequence
- •Agent-summarized conversation context (for example, previously successful setup steps)
Output conventions
- •Primary scripts go in
scripts/and use*.sh.tsnaming. - •Include shebang:
#!/usr/bin/env bun. - •Prefer shared helpers from
scripts/helpers/instead of duplicating utility logic. - •Default to explicit, readable command flow and fail fast on command errors.
- •For external docs scraping workflows, default output to split per-page
.ext.mdfiles and include anindex.ext.mdin the target directory.
Implementation workflow
- •Identify script intent, inputs, and expected outputs from provided context.
- •If input comes from troubleshooting history, separate successful steps from failed or diagnostic attempts.
- •Exclude failed retries, dead-end commands, and one-off debugging probes from final reusable script logic.
- •Choose script name and location under
scripts/. - •Reuse or add helper utilities in
scripts/helpers/when needed. - •Implement Bun-based script and run it once when safe to verify behavior.
- •Update docs (
README.mdorAGENTS.md) when conventions or entrypoints change.
Context capture rules
- •When converting conversation history into a script, include only the final successful sequence.
- •Keep troubleshooting commands as optional notes in output, not in executable script flow.
- •If multiple working paths exist, prefer the shortest deterministic path.
Notes
- •Keep scripts idempotent where practical.
- •Avoid embedding secrets or environment-specific absolute paths.
- •If verification cannot be run safely, state exactly what should be executed by the user.