English · Español
Welcome. This project is Software as a Medical Device (SaMD) Class {{SAMD_CLASS}} under IEC 62304 + ISO 14971. The change process is controlled and is regulatory evidence (IEC 62304 §5/§8): it isn't bureaucracy, it's what makes certification possible. This page is the front door; the detail lives in the documents linked below.
- No one commits or pushes without the owner's explicit OK ({{OWNER}}). Before each commit, report which files go up and wait for authorization. Destructive operations (
reset --hard,push --force, deleting branches) require a separate OK. - Green tests with reported numbers. Nothing is declared "done" without running the tests tied to the change and reporting the results (SaMD §5.7). "Renders without crashing" is not a valid test.
- The user never sees a traceback. On-screen errors: empathetic messages + the correct HTTP code. A user in crisis must not see "500 Internal Server Error".
- Traceability in the SAME PR. If you touch a clinical algorithm, a schema, a business rule, or a security flow, you update the DHF (TECHNICAL_DEBT_SUMMARY + Master Map + Risk Matrix / TRACEABILITY when applicable) in the same PR, not in a separate sprint.
- Identity from the token only. Never accept
user_idfrom the body, query, or headers — only from the decoded token. - Impact analysis before declaring "fixed". Review ALL consumers of the symbol you touched (global
grep), not just the file where the bug was reported.
1. Branch from your default branch: git switch -c feat/my-change
2. Implement + tests (delegate to the agents in .claude/agents/ when the layer fits)
3. Local CI green: bash scripts/run_local_ci.sh
4. Report the files to commit and wait for the owner's OK
5. Open the PR; CI runs lint + types + tests + contract drift + traceability
6. Before closing a large block: bash scripts/audit_project_state.sh
- Passes linting + local CI (
scripts/run_local_ci.sh). - Tied tests green, numbers reported.
- Technical debt reduced by at least one point (Boy Scout rule).
- Documentation up to date (Master Map, TECHNICAL_DEBT, TRACEABILITY, and Risk Matrix when applicable).
- Accessible UI (WCAG AA contrast, flat design, canonical modals).
- Change communicated to the owner with a clear rationale and explicitly authorized.
CLAUDE.md— the agent ruleset core: Rule 0, how to work with the owner, empathy, boy scout, impact analysis and block closure. Loads on every turn..claude/rules/— the rest of the ruleset: four path-scoped rules that load only when that layer is touched (backend,frontend-ux,testing,github-actions) plus four cross-cutting ones (orquestacion,agentes,comandos,sesiones-paralelas). A new layer rule goes here; a rule that always applies goes in the core..agents/workflows/protocolo_desarrollo.md— the stable, agent-agnostic development process.docs/03_software_development_plan/DEVELOPMENT_GUIDE_COMPLETE.md— the complete guide: setup, commands, standards, PR flow.docs/03_software_development_plan/COMPLETE_TESTING_STRATEGY.md— the SaMD-grade verification strategy.docs/03_software_development_plan/AUDIT_PROTOCOL.md— how a changeset/phase is audited against SaMD.docs/05_design_decisions/RFC-TEMPLATE.md— template for proposing a structural decision.
Vulnerabilities are NOT reported in public issues. Contact {{OWNER}} privately. See docs/07_regulatory_and_compliance/INCIDENT_RESPONSE_PLAN.md.