BlackBread
The fill-in form of the *Execution prompt contract* (`execution-contract.md` §1) and the *Required slice contract* (`architecture-planning.md`). It restates no rationale — it is the skeleton to fill.
Install / Use
npx skills add carlitotate12160-tech/BlackBreadInstalls into whichever agent you are using.
Devin Config
Devin AI agent config
Quality Score
Category
LegalSupported Platforms
Skill content
View source on GitHubExecution Prompt Template
The fill-in form of the Execution prompt contract (execution-contract.md §1) and the
Required slice contract (architecture-planning.md). It restates no rationale — it is the
skeleton to fill. Copy it, replace every <...>, delete the parenthetical guidance, and state each
rule once in its owning section (no rule repeated across sections). One slice per prompt.
Use the blackbread-engineering skill in <IMPLEMENT | FIX | REVIEW> mode.
<Milestone>.<slice-id> — <one-sentence outcome>
1. Outcome and scope
- SLICE:
M<n>.<id>— <what this slice makes true, one sentence>. - MODE:
<IMPLEMENT | FIX | REVIEW>. One bounded slice; you are the sole implementation owner of the branch. Do not re-plan or rescope already-accepted architecture; do not ask another agent to edit the branch.
2. Baseline (verify live; the checkpoint below is for drift detection only)
- protected base:
<sha>PR head:<sha or n/a>branch:<name> - Read completely:
AGENTS.md,.github/agent-delivery.json,ENGINEERING-STATE.md, and only the implementation/test files this slice requires. - If live base, head, diff, authority, open gaps, or blocker set differs materially from the
checkpoint:
STOP/SPLITwithout editing and report the drift.
3. File map (the only files that may change)
| File | Action | Responsibility | Expected size |
| ---- | ------ | -------------- | ------------- |
| <path> | <add | modify> | <one responsibility> | <small | medium> |
| tests/<path> | modify | <proof it carries> | <small> |
All other files are forbidden. If a change outside this map appears necessary, STOP/SPLIT and
report the violated assumption — do not implement it.
4. Requirement / capability context
- REQ/CAP IDs touched:
<ENG-/REC-/STR-/CAP-/... + capability-registry IDs>— state each one's current state (DECIDED | IMPLEMENTED | VERIFIED | RELEASED). - Why, and what it connects to: <one short paragraph>. Moat = composition + proof, not API-wrapping. Candidate != authorization — every destination still passes the deterministic gate downstream.
5. Invariants and non-goals (stated once, here)
- Preserve: <deterministic safety = pure code, LLM emits typed proposals only; atomic transaction; lock/serialization point; tenant and engagement isolation; RLS fail-closed; immutable provenance/lineage; sealed preimages, migrations, golden vectors, versions; open-gap honesty>.
- Do NOT: <change production code or schema unless this slice is that change; introduce a new trust boundary; scope beyond this slice; rename public or private runtime symbols; move policy into tests/migrations/config to dodge budgets; density-game (compress, strip docs, single-letter names, merge responsibilities); open a second PR; force-push or rewrite history; call CodeRabbit, Sourcery, Codex, or any advisory bot manually; merge>.
6. Proof obligations (RED first; one oracle per claim)
For each behavior being added or corrected:
<claim / defect name> — <test name>
- Passes only if: <exact positive conditions, one per line>.
- Must NOT pass because: <a broad
Exception; asleep/timing-luck; a tautological count; RLS concealing invalid data from the observer; an unrelated failure>. - RED evidence: introduce a temporary, non-committed mutation or seam that removes the property, show the strengthened test fails, then restore the real implementation before GREEN.
Wiring proof (non-island) — <test name>
- Drive the real Conductor autonomous path (not a fixture or a
live_firerunner) and assert the new capability's output actually reaches its downstream consumer. Register the wiring transition in the wiring-debt gate if the repository tracks one.
7. Budget (single source)
- Caps come from
config/quality-budgets.json(module <= 400, function <= 50, McCabe <= 10) and the protected-base allowance; do not restate numbers, do not raise a cap. - Predict the runtime line/file delta with review margin. Budget is an architecture signal, not a
formatting target — if honest code will not fit,
STOP/SPLIT, do not compress.
8. STOP/SPLIT conditions (the single list; other sections reference this)
Stop without broadening the diff and report exact evidence when: <live base/head drift; a further implementation responsibility becomes necessary; production behavior or a sealed migration must change; the proof needs a new runtime seam; a real-PostgreSQL proof is unavailable; an authority conflict appears; a module cannot stay in budget without compression; a new correctness-invariant class appears>.
9. Local preflight (ordered; opening the PR before this is green is a task failure)
- focused RED or temporary-mutation proof;
- focused corrected/added tests GREEN;
- affected compatibility suites (e.g. ledger, tenancy);
- real PostgreSQL for migration/transaction/lock/RLS/replay claims;
uv sync --locked --all-groups;make check;- size / diff / governance checks;
git diff --check;- inspect the complete diff against protected
main.
Oracle ARM64 only (.venv312/bin/python3 -m pytest or make check); never bare pytest. git pull
and re-verify first.
10. Delivery
- One conventional commit; normal fast-forward push to the existing branch; no force-push; no new PR; no manual advisory-bot call; let configured CI and the binding PR-Agent/DeepSeek review run; do not merge.
- Inspect every thread against the exact new head. Fix a reproduced valid finding; for a stale or false-positive finding, reply with exact-head evidence and resolve its thread with that note. Dispositioning advisory findings is a practice that satisfies thread resolution, not a blocking gate; only a reproduced valid correctness/security finding blocks. Resolve a thread only after its evidence exists.
11. Required return
verified base before editing; old head; new head; complete changed-file list; RED/mutation evidence;
focused GREEN evidence; real-PostgreSQL evidence; full make check, coverage, and safety-coverage
results; quality and diff-budget results; disposition for every previously unresolved thread; current
CI status; current binding PR-Agent exact-head review status; unresolved threads remaining; open gaps;
claims explicitly not made; MERGEABLE or NOT MERGEABLE; exactly one next action. Do not merge.
Related Skills
OpenMetadata
15.1kThe Open Context Layer for Data and AI , OpenMetadata is the open platform for building trusted data context and business semantics for humans, AI assistants, and agents.
tabularis
4.6kOpen-source desktop SQL workspace for PostgreSQL, MySQL/MariaDB, SQLite and 15+ more databases like DuckDB, ClickHouse, Redis and Firestore. Built-in MCP server for Claude, Cursor and Devin, SQL notebooks and visual EXPLAIN.
pyspark-etl-best-practices-cursorrules-prompt-file
40.7kCursor rules for PySpark ETL development with code style, joins, window functions, map operations, and Iceberg patterns.
claude-ads
8.6kClaude-first paid-media operations skill for Claude Code across 12 ad platforms (Google, Meta, YouTube, LinkedIn, TikTok, Microsoft, Apple, Amazon, Reddit, Pinterest, Snapchat, X): source-grounded audits, deterministic scoring, versioned JSON reports, and capability-gated account changes.
Security Score
Audited on Invalid Date
