Development context receipt¶

Sources and pins¶

• addyosmani/agent-skills commit be4e44a9fbc5e8df0beaefadbb28bd22ee61cc39, MIT license:
• skills/context-engineering/SKILL.md
• skills/source-driven-development/SKILL.md
• Official OpenAI documentation read on 2026-09-16:
• https://learn.chatgpt.com/docs/build-skills
• https://learn.chatgpt.com/docs/agent-configuration/agents-md

Both pinned upstream skill files were read completely. They are referenced as guidance and are not copied into this repository.

Context selected for initial implementation¶

Initial orientation required the seven authoritative planning artifacts and the two upstream skill files: 365,584 bytes total, roughly 91,000 tokens at four bytes per token. That one-time load retained the product identity, reuse gate, privacy/mutation limits, command contracts, release criteria, and historical evidence boundaries.

Routine tasks use the smaller chain in this repository:

1. AGENTS.md for durable boundaries and commands;
2. .agents/skills/shoulda-development/SKILL.md for task routing;
3. one applicable architecture contract or prior-art receipt;
4. the exact source/schema/test files being changed; and
5. current error output only when a check fails.

The initial repository guidance plus public implementation and reuse contracts are under 15 KiB. No missing-context rework was observed during the foundation task. If a later task misses an obligation, this receipt records the omitted file and rework before any compression tool is considered.

Decisions changed by the guidance¶

• The long planning corpus became a one-time orientation input rather than permanent prompt material.
• Dependency and API choices were refreshed from current primary metadata before code was written.
• The repository skill links to contracts rather than duplicating them.
• Fetched documentation is treated as evidence data, never as authority to expand scope or perform unrelated actions.
• Unknown, stale, conflicting, and unverified facts remain explicit in both product records and development receipts.

Guidance deliberately not applied¶

• The upstream examples use several agent-specific filenames; this repository uses Codex-native AGENTS.md and .agents/skills documented by OpenAI.
• Context-size heuristics are observations, not hard product limits. Correctness and required evidence take priority.
• Model/context tooling named by upstream guidance is not installed because the deterministic selection packet is currently within budget.