17 added, 49 removed. Audit A to A.
# AI work instructions
- Before changing Hope:
-
- - Read [PRINCIPLES.md](PRINCIPLES.md).
- - Follow the workflow in [CONTRIBUTING.md](CONTRIBUTING.md).
- - Read [docs/architecture.md](docs/architecture.md) before changing
- implementation dependencies, shared code or assets, a main folder, a build
- boundary, or delivery structure.
- - Read the matching `plugins/hope/skills/<feature>/SKILL.md` before changing
- feature behavior.
- - Read [docs/design.md](docs/design.md) before changing a Hope GUI.
-
- ## Writing
-
- Use the Hope Write Skill whenever clearer language would improve the work.
-
- Write applies the shared
- [writing standard](plugins/hope/skills/write/references/writing-standard.md)
- inside the active task.
-
- When implementation or another Skill owns the task, Write remains
- cross-cutting. It does not create another workflow, worker, result, scope, or
- completion condition.
-
- ## Completion
-
- Before finishing, review the full changed scope against
- [Prefer simple, direct design](PRINCIPLES.md#prefer-simple-direct-design).
-
- Remove support files, generation, packaging, documentation, and tests that no
- longer serve the remaining product.
-
- Link to one authority instead of keeping parallel behavior descriptions unless
- another file owns a distinct contract or obligation.
-
- Verify changed product behavior.
-
- Skill discovery and manifest validity alone are not evidence that behavior
- works.
+ Follow [CONTRIBUTING.md](CONTRIBUTING.md) for required reading, verification,
+ release preparation, and submission.
- Run the relevant checks from [CONTRIBUTING.md](CONTRIBUTING.md) and report any
- remaining verification gap.
+ Carry an action request through to its intended result. Use the conversation's
+ existing decisions and authorization; choose routine, reversible details
+ without asking again. Ask only when missing information changes the result or
+ an action needs authority the person has not given. Continue independent,
+ authorized work while an answer is pending.
- Before completing any file-changing task, follow the release decision and
- preparation steps in [docs/release.md](docs/release.md), then run
- `npm run check`.
+ Explicit user instructions take precedence over Skill guidance. If a Skill
+ would pause or redirect authorized work, first check whether that rule applies.
+ When it does, quote and link the exact instruction and explain the remaining
+ decision. Prepare the authorized work so any needed approval concerns a
+ concrete, reviewable result.
- Before committing or opening a pull request, choose one final title that follows
- [the contribution rule](CONTRIBUTING.md#record-and-submit-the-change). Validate
- it with `npm run check:title -- "<title>"`, then use it for both the commit and
- pull request.
+ Use [Hope Write](plugins/hope/skills/write/SKILL.md) wherever clearer language
+ helps. Keep updates and results concise; report the change, relevant evidence,
+ and any verification gap.
- Do not finish while `npm run check` fails. Do not commit, open a pull request,
- or push while the selected title fails `npm run check:title -- "<title>"`.
+ These collaboration rules draw on the
+ [GPT-6 Astra prompting guide](https://developers.openai.com/api/docs/guides/latest-model?model=gpt-6-astra#prompting-best-practices).