Skip to content
Skills Library
Programming · Planning

ai-safe2-development-method

/ai-safe2-development-method

Plan and execute material repository changes with the AI SAFE2 development method. Use when a coding task needs a delivery-shape decision, risk-adjusted design depth, isolated implementation, test-first or explicit substitute evidence, systematic debugging, review cadence, before/after comparison, or a completion receipt. Do not use for read-only explanations, trivial edits, or as authority to merge, release, deploy, accept risk, or change policy.

Licence
MIT
Last repository change
23.10.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
5
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 23.10.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

Plan and execute material repository changes with the AI SAFE2 development method. Use when a coding task needs a delivery-shape decision, risk-adjusted design depth, isolated implementation, test-first or explicit substitute evidence, systematic debugging, review cadence, before/after comparison, or a completion receipt. Do not use for read-only explanations, trivial edits, or as authority to merge, release, deploy, accept risk, or change policy.

Original author description

Plan and execute material repository changes with the AI SAFE2 development method. Use when a coding task needs a delivery-shape decision, risk-adjusted design depth, isolated implementation, test-first or explicit substitute evidence, systematic debugging, review cadence, before/after comparison, or a completion receipt. Do not use for read-only explanations, trivial edits, or as authority to merge, release, deploy, accept risk, or change policy.

How it works

  1. Workflow
  2. Decision Rules
  3. References
Original SKILL.md headings · 50 lines, 382 words
SKILL.md preview
---
name: ai-safe2-development-method
description: Plan and execute material repository changes with the AI SAFE2 development method. Use when a coding task needs a delivery-shape decision, risk-adjusted design depth, isolated implementation, test-first or explicit substitute evidence, systematic debugging, review cadence, before/after comparison, or a completion receipt. Do not use for read-only explanations, trivial edits, or as authority to merge, release, deploy, accept risk, or change policy.
---

# AI SAFE2 Development Method

Convert software-development discipline into decision evidence. Keep policy,
implementation, verification, semantic review, and human authority separate.

## Workflow

1. Read the nearest repository instructions and current policy before acting.
2. Classify delivery shape independently from risk:
   - `spike`: disposable learning; implementation does not become production output.
   - `bounded`: limited scope with explicit acceptance conditions and exclusions.
   - `architectural`: cross-boundary or durable design requiring a written spec and plan.
3. Create a `safe2.development-plan-source.v1` source and run
   `safe2 dev plan SOURCE --output PLAN`. Use the repository policy when present.
4. If the plan is `review_required`, pause only for the named missing decision.
   Low- and medium-risk bounded work may proceed under existing user authority.
   Never work around an `invalid` plan.
5. Isolate repository-changing work. Preserve unrelated user changes.
6. Use the selected evidence mode:
   - code: observe a failing behavior first, then the passing behavior;
   - existing ambiguous behavior: characterize before changing it;
   - contracts/configuration: establish the invalid/valid boundary first;
   - documents/UI: use render or interaction validation;
   - spike: record the hypothesis, result, and discard decision.
7. When a check fails, localize the cause before proposing a fix. Do not stack
   speculative changes. Re-run the smallest affected check after each fix.
8. Review at the cadence in the plan. Treat PR-Agent, Greptile, and other model
   reviewers as attributed evidence. Provider failure is `unavailable`, not pass.
9. Capture comparable before/after evidence when required. If no safe baseline
   exists, preserve that gap and add a regression test without claiming a direct delta.
10. Before any completion claim, run fresh verification against the final revision
    and produce `safe2 dev receipt PLAN SOURCE --artifact-root ROOT --output RECEIPT`.

## Decision Rules

- Delivery shape controls design depth; risk controls scrutiny and authority.
- Architectural work and high/critical risk require explicit human design approval.
- Critical risk requires a threat model and specialist review cadence.
- A spike is disposable. Reusable output requires a bounded or architectural plan.
- Deterministic checks remain authoritative. Semantic reviewers explain and
  corroborate; they do not replace tests, schemas, scanners, or human decisions.
- Never infer permission to merge, release, deploy, change policy, access secrets,
  or accept residual risk from a plan or receipt.

## References

- `references/method.md`: classification, debugging, evidence, and receipt guidance
- `references/evaluation.md`: pressure cases and regression expectations
Files in this package (5)
  • SKILL-CARD.md 1 KB
  • SKILL.md 3 KB
  • agents/openai.yaml 1 KB
  • references/evaluation.md 1 KB
  • references/method.md 2 KB

Versions

The database has no other skills named ai-safe2-development-method.

Database snapshot: 13.11.2026 · package version 5a96f45f8f3e (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100delivery-method
    Coordinate analysis, debugging, features and complete documentation through grounded specifications, adversarial plans, verified commits and authorized merge. TRIGGER WHEN: the user wants to brainstorm or design a project, define specifications or implementation plans,…
  • 100test-driven-development
    Use when implementing any feature or bugfix before implementation changes. Enforces test-first behavior: write a failing test, verify failure, implement minimal code, and prove green.
  • 90systematic-debugging
    Plan, implement, or review systematic debugging work in an existing codebase with compatibility, security, and verification controls. Use when the user explicitly requests systematic debugging work.
  • 100finishing-a-development-branch
    Use when implementation is complete, all tests pass, and you need to decide how to integrate the work - guides completion of development work by presenting structured options for merge, PR, or cleanup
  • 100david-workflow
    End-to-end workflow for a feature, fix or change. Plan in phases with a verification plan up front, implement, verify for real and review with sub agents, open a PR (or a gh stack) with evidence attached, drive CI and review comments to green, test on a preview deploy with a…
  • 100coding-discipline
    Lightweight coding discipline for AI coding agents: surface assumptions before coding, keep changes small and surgical, write failing tests first, verify outcomes. Use when writing, reviewing, refactoring, or debugging non-trivial code. Do not use for trivial one-line edits,…

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.