Skip to content

12_repo_quality_gate_command_matrix.md

Latest commit

 

History

History
163 lines (113 loc) · 6.45 KB

12_repo_quality_gate_command_matrix.md

File metadata and controls

163 lines (113 loc) · 6.45 KB

Repo Quality Gate Command Matrix

Purpose

The packet needs explicit gate commands so "all green" is operational rather than interpretive.

This file is the command source of truth for every implementation wave.

Exact Command Sets

Use the exact command set named in the repo matrix.

Repeated command sets are intentional. The packet is removing operator guesswork, not deduplicating prose.

ROOT_FULL

Run at repo root in this order:

  1. mix format
  2. mix compile --warnings-as-errors
  3. mix test
  4. mix credo --strict
  5. mix dialyzer
  6. mix docs

ROOT_NO_STATIC_ANALYSIS

Run at repo root in this order:

  1. mix format
  2. mix compile --warnings-as-errors
  3. mix test
  4. mix docs

Use this only for repos that do not currently advertise mix credo and mix dialyzer.

ROOT_CI_ALIAS

Run the repo-owned mix ci alias. For execution_plane, this expands to:

  1. mix format --check-formatted
  2. mix compile --warnings-as-errors
  3. MIX_ENV=test mix test
  4. mix credo --strict
  5. MIX_ENV=test mix dialyzer --force-check
  6. mix docs --warnings-as-errors

ROOT_NO_DOCS

Run at repo root in this order:

  1. mix format
  2. mix compile --warnings-as-errors
  3. mix test
  4. mix credo --strict
  5. mix dialyzer

Use this only for repos that do not currently advertise mix docs.

MONOREPO_FULL

Run at repo root in this order:

  1. mix monorepo.format
  2. mix monorepo.compile
  3. mix monorepo.test
  4. mix monorepo.credo --strict
  5. mix monorepo.dialyzer
  6. mix monorepo.docs

ARCHIVAL_ROOT

Start from ROOT_FULL.

If a wave reduces the repo to an archival shell and removes tasks or code paths, rerun every task that still exists and record each skipped task explicitly as not_applicable_after_archival_shell_reduction in the checklist evidence section.

Repo Matrix

Repo Exact gate stack Notes
execution_plane ROOT_CI_ALIAS root mix ci is the Blitz workspace gate; publishable packages still own their package-local mix ci aliases
external_runtime_transport ARCHIVAL_ROOT use full root stack while code still exists; record any archival-shell task removals explicitly
pristine MONOREPO_FULL use the root mix monorepo.* aliases
prismatic MONOREPO_FULL use the root mix monorepo.* aliases
notion_sdk ROOT_FULL root repo commands
github_ex ROOT_FULL root repo commands
linear_sdk ROOT_FULL mix ci is stronger but the packet source of truth remains the explicit root stack above
reqllm_next ROOT_FULL the repo also has mix quality; the packet source of truth remains the explicit root stack above
cli_subprocess_core ROOT_FULL root repo commands
codex_sdk ROOT_FULL root repo commands
claude_agent_sdk ROOT_FULL root repo commands
gemini_cli_sdk ROOT_FULL root repo commands
amp_sdk ROOT_FULL root repo commands
agent_session_manager ROOT_FULL root repo commands
self_hosted_inference_core ROOT_FULL root repo commands
llama_cpp_sdk ROOT_FULL root repo commands
jido_integration MONOREPO_FULL use the root mix monorepo.* aliases
citadel ROOT_FULL Brain kernel; authority compilation and AuthorityDecision.v1 author
packet_docs packet-local validation commands run reference scan, stale-name scan, metadata alignment scan, and Elixir parse checks

Packet-Local Validation Commands

For packet_docs, run:

  1. packet-local markdown reference scan:

    elixir -e 'root = File.cwd!(); files = Path.wildcard("**/*.md"); regex = ~r/`((?:prompts|technical|adrs|JIDO_BRAIN_CONTRACT_CONTEXT)\/[^`]+?\.md)`/; missing = for file <- files, {:ok, text} = File.read(file), [_, rel] <- Regex.scan(regex, text), not File.exists?(Path.join(root, rel)) do {file, rel} end; Enum.each(missing, fn {file, rel} -> IO.puts("MISSING_REF #{file} -> #{rel}") end); IO.puts("missing_count=#{length(missing)}")'

  2. stale-name or stale-path scan:

    rg -n "03_minimal_viable_http_and_process_lanes_implementation_prompt|04_provider_and_dependent_family_adoption_of_minimal_lanes_implementation_prompt|05_session_bearing_lane_convergence_implementation_prompt|06_durable_truth_replay_approval_and_identity_alignment_implementation_prompt|08_active_owner_retirement_and_surface_cleanup_implementation_prompt|09_cross_repo_conformance_and_failure_hardening_implementation_prompt|10_final_verification_and_doc_sync_prompt|Workstream" README.md atlas.md technical adrs prompts JIDO_BRAIN_CONTRACT_CONTEXT brain_spine_execution_plane_architecture_review.md --glob '!technical/12_repo_quality_gate_command_matrix.md'

    The expected clean result is no matches.

  3. packet metadata alignment scan:

    elixir prompts/validate_packet_metadata.exs

    The expected clean result is packet_metadata_ok.

    This scan must verify:

    • phase naming alignment across prompts.txt and runner_config.exs
    • prompt-file existence
    • primary target coverage in prompts/commit-message.txt
    • valid runner target names for every commit section
    • optional commit-section scope limited to named conditional upstream repos declared in the prompt

    If ADR-016 forces a new upstream repo outside the currently named conditional scope, amend the wave prompt, checklist, and prompts/commit-message.txt first, then rerun this scan.

  4. Code.string_to_quoted!/1 parse checks for prompts/runner_config.exs, prompts/run_prompts.exs, and prompts/validate_packet_metadata.exs:

    elixir -e 'for f <- ["prompts/runner_config.exs", "prompts/run_prompts.exs", "prompts/validate_packet_metadata.exs"] do File.read!(f) |> Code.string_to_quoted!() end; IO.puts("elixir_parse_ok")'

Equivalent commands are allowed if they provide the same evidence and are recorded in the checklist.

Partial Failure Rule

If an upstream fix is needed:

  • fix forward in the upstream repo during the same wave
  • rerun the relevant gate stack in that upstream repo
  • rerun the relevant gate stack in the impacted downstream repo
  • record both runs in the checklist evidence section

Closure Evidence Rule

Each wave checklist must record at least:

  • repos actually touched
  • upstream repos touched beyond the primary target list
  • exact commands run
  • command outcomes or summary evidence
  • whether capability-local owner retirement happened in the same wave
  • any substitutions for repo-local equivalent commands