@apralabs/apra-fleet 0.4.2 → 0.4.3
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +38 -5
- package/dist/agents/backlog-groomer.md +25 -0
- package/dist/agents/ci-watcher.md +11 -0
- package/dist/agents/deployer.md +37 -2
- package/dist/agents/doer.md +36 -12
- package/dist/agents/harvester.md +11 -0
- package/dist/agents/integ-test-runner.md +11 -0
- package/dist/agents/kb-reconciler.md +9 -0
- package/dist/agents/plan-reviewer.md +110 -22
- package/dist/agents/planner.md +125 -37
- package/dist/agents/regression-test-runner.md +11 -0
- package/dist/agents/reviewer.md +13 -2
- package/dist/agents/schemas/plan-reviewer-output.json +37 -0
- package/dist/cli/agent-transform.d.ts +83 -2
- package/dist/cli/agent-transform.d.ts.map +1 -1
- package/dist/cli/agent-transform.js +266 -21
- package/dist/cli/agent-transform.js.map +1 -1
- package/dist/cli/auth.d.ts.map +1 -1
- package/dist/cli/auth.js +28 -9
- package/dist/cli/auth.js.map +1 -1
- package/dist/cli/config.d.ts +7 -5
- package/dist/cli/config.d.ts.map +1 -1
- package/dist/cli/config.js +44 -34
- package/dist/cli/config.js.map +1 -1
- package/dist/cli/install.d.ts +33 -0
- package/dist/cli/install.d.ts.map +1 -1
- package/dist/cli/install.js +290 -23
- package/dist/cli/install.js.map +1 -1
- package/dist/cli/kb-directives.d.ts.map +1 -1
- package/dist/cli/kb-directives.js +2 -1
- package/dist/cli/kb-directives.js.map +1 -1
- package/dist/cli/register-member.d.ts.map +1 -1
- package/dist/cli/register-member.js +6 -2
- package/dist/cli/register-member.js.map +1 -1
- package/dist/commands/kb-server.d.ts +3 -0
- package/dist/commands/kb-server.d.ts.map +1 -1
- package/dist/commands/kb-server.js +43 -3
- package/dist/commands/kb-server.js.map +1 -1
- package/dist/index.js +3 -3
- package/dist/os/linux.d.ts +6 -0
- package/dist/os/linux.d.ts.map +1 -1
- package/dist/os/linux.js +35 -2
- package/dist/os/linux.js.map +1 -1
- package/dist/os/os-commands.d.ts +61 -2
- package/dist/os/os-commands.d.ts.map +1 -1
- package/dist/os/windows-gitbash.d.ts +5 -0
- package/dist/os/windows-gitbash.d.ts.map +1 -1
- package/dist/os/windows-gitbash.js +15 -0
- package/dist/os/windows-gitbash.js.map +1 -1
- package/dist/os/windows.d.ts +6 -0
- package/dist/os/windows.d.ts.map +1 -1
- package/dist/os/windows.js +46 -3
- package/dist/os/windows.js.map +1 -1
- package/dist/providers/agy.d.ts +115 -6
- package/dist/providers/agy.d.ts.map +1 -1
- package/dist/providers/agy.js +569 -53
- package/dist/providers/agy.js.map +1 -1
- package/dist/providers/claude.d.ts +81 -3
- package/dist/providers/claude.d.ts.map +1 -1
- package/dist/providers/claude.js +344 -10
- package/dist/providers/claude.js.map +1 -1
- package/dist/providers/codex.d.ts +4 -2
- package/dist/providers/codex.d.ts.map +1 -1
- package/dist/providers/codex.js +19 -2
- package/dist/providers/codex.js.map +1 -1
- package/dist/providers/copilot.d.ts +4 -2
- package/dist/providers/copilot.d.ts.map +1 -1
- package/dist/providers/copilot.js +20 -2
- package/dist/providers/copilot.js.map +1 -1
- package/dist/providers/none.d.ts +4 -2
- package/dist/providers/none.d.ts.map +1 -1
- package/dist/providers/none.js +17 -2
- package/dist/providers/none.js.map +1 -1
- package/dist/providers/opencode.d.ts +3 -1
- package/dist/providers/opencode.d.ts.map +1 -1
- package/dist/providers/opencode.js +22 -2
- package/dist/providers/opencode.js.map +1 -1
- package/dist/providers/provider.d.ts +201 -6
- package/dist/providers/provider.d.ts.map +1 -1
- package/dist/providers/provider.js +56 -0
- package/dist/providers/provider.js.map +1 -1
- package/dist/services/agy-project.d.ts +102 -0
- package/dist/services/agy-project.d.ts.map +1 -0
- package/dist/services/agy-project.js +356 -0
- package/dist/services/agy-project.js.map +1 -0
- package/dist/services/auth-socket.js +4 -4
- package/dist/services/auth-web.js +1 -1
- package/dist/services/cloud/lifecycle.d.ts.map +1 -1
- package/dist/services/cloud/lifecycle.js +33 -4
- package/dist/services/cloud/lifecycle.js.map +1 -1
- package/dist/services/credential-cleanup.d.ts.map +1 -1
- package/dist/services/credential-cleanup.js +35 -17
- package/dist/services/credential-cleanup.js.map +1 -1
- package/dist/services/exit-drain.d.ts +44 -0
- package/dist/services/exit-drain.d.ts.map +1 -0
- package/dist/services/exit-drain.js +52 -0
- package/dist/services/exit-drain.js.map +1 -0
- package/dist/services/github-app.d.ts.map +1 -1
- package/dist/services/github-app.js +32 -4
- package/dist/services/github-app.js.map +1 -1
- package/dist/services/http-transport.d.ts +1 -0
- package/dist/services/http-transport.d.ts.map +1 -1
- package/dist/services/http-transport.js +48 -1
- package/dist/services/http-transport.js.map +1 -1
- package/dist/services/knowledge/kb-config.d.ts +42 -0
- package/dist/services/knowledge/kb-config.d.ts.map +1 -0
- package/dist/services/knowledge/kb-config.js +54 -0
- package/dist/services/knowledge/kb-config.js.map +1 -0
- package/dist/services/knowledge/kb-providers.d.ts +7 -1
- package/dist/services/knowledge/kb-providers.d.ts.map +1 -1
- package/dist/services/knowledge/kb-providers.js +191 -17
- package/dist/services/knowledge/kb-providers.js.map +1 -1
- package/dist/services/knowledge/require-sqlite-project.d.ts +32 -0
- package/dist/services/knowledge/require-sqlite-project.d.ts.map +1 -0
- package/dist/services/knowledge/require-sqlite-project.js +38 -0
- package/dist/services/knowledge/require-sqlite-project.js.map +1 -0
- package/dist/services/model-pricing.d.ts.map +1 -1
- package/dist/services/model-pricing.js +7 -3
- package/dist/services/model-pricing.js.map +1 -1
- package/dist/services/secret-token.d.ts +38 -0
- package/dist/services/secret-token.d.ts.map +1 -0
- package/dist/services/secret-token.js +51 -0
- package/dist/services/secret-token.js.map +1 -0
- package/dist/services/service-manager/types.d.ts +3 -1
- package/dist/services/service-manager/types.d.ts.map +1 -1
- package/dist/services/service-manager/windows.d.ts +14 -2
- package/dist/services/service-manager/windows.d.ts.map +1 -1
- package/dist/services/service-manager/windows.js +65 -7
- package/dist/services/service-manager/windows.js.map +1 -1
- package/dist/services/ssh.d.ts.map +1 -1
- package/dist/services/ssh.js +58 -2
- package/dist/services/ssh.js.map +1 -1
- package/dist/services/stall/read-log-tail.d.ts.map +1 -1
- package/dist/services/stall/read-log-tail.js +20 -10
- package/dist/services/stall/read-log-tail.js.map +1 -1
- package/dist/services/stall/stall-detector.d.ts +53 -4
- package/dist/services/stall/stall-detector.d.ts.map +1 -1
- package/dist/services/stall/stall-detector.js +212 -34
- package/dist/services/stall/stall-detector.js.map +1 -1
- package/dist/services/strategy.d.ts.map +1 -1
- package/dist/services/strategy.js +79 -2
- package/dist/services/strategy.js.map +1 -1
- package/dist/services/substitution-engine.js +2 -2
- package/dist/services/substitution-engine.js.map +1 -1
- package/dist/services/tool-registry.d.ts.map +1 -1
- package/dist/services/tool-registry.js +5 -3
- package/dist/services/tool-registry.js.map +1 -1
- package/dist/services/vcs/azure-devops.d.ts +66 -1
- package/dist/services/vcs/azure-devops.d.ts.map +1 -1
- package/dist/services/vcs/azure-devops.js +192 -21
- package/dist/services/vcs/azure-devops.js.map +1 -1
- package/dist/services/vcs/types.d.ts +1 -1
- package/dist/skills/fleet-sprint-cli/SKILL.md +18 -0
- package/dist/skills/pm/SKILL.md +1 -1
- package/dist/skills/pm/cost.md +0 -3
- package/dist/skills/pm/fleet-addendum.md +2 -2
- package/dist/tools/check-status.d.ts +7 -4
- package/dist/tools/check-status.d.ts.map +1 -1
- package/dist/tools/check-status.js +2 -0
- package/dist/tools/check-status.js.map +1 -1
- package/dist/tools/compose-permissions.d.ts +6 -0
- package/dist/tools/compose-permissions.d.ts.map +1 -1
- package/dist/tools/compose-permissions.js +182 -34
- package/dist/tools/compose-permissions.js.map +1 -1
- package/dist/tools/credential-store-set.js +1 -1
- package/dist/tools/execute-command.d.ts.map +1 -1
- package/dist/tools/execute-command.js +31 -20
- package/dist/tools/execute-command.js.map +1 -1
- package/dist/tools/execute-prompt.d.ts +15 -2
- package/dist/tools/execute-prompt.d.ts.map +1 -1
- package/dist/tools/execute-prompt.js +1196 -892
- package/dist/tools/execute-prompt.js.map +1 -1
- package/dist/tools/kb-capture.d.ts +2 -2
- package/dist/tools/kb-capture.d.ts.map +1 -1
- package/dist/tools/kb-capture.js +29 -5
- package/dist/tools/kb-capture.js.map +1 -1
- package/dist/tools/kb-export.d.ts +2 -2
- package/dist/tools/kb-export.d.ts.map +1 -1
- package/dist/tools/kb-export.js +3 -28
- package/dist/tools/kb-export.js.map +1 -1
- package/dist/tools/kb-feedback.d.ts.map +1 -1
- package/dist/tools/kb-feedback.js +3 -1
- package/dist/tools/kb-feedback.js.map +1 -1
- package/dist/tools/kb-freshness-sweep.d.ts.map +1 -1
- package/dist/tools/kb-freshness-sweep.js +3 -1
- package/dist/tools/kb-freshness-sweep.js.map +1 -1
- package/dist/tools/kb-import.d.ts.map +1 -1
- package/dist/tools/kb-import.js +2 -1
- package/dist/tools/kb-import.js.map +1 -1
- package/dist/tools/kb-list.d.ts.map +1 -1
- package/dist/tools/kb-list.js +3 -1
- package/dist/tools/kb-list.js.map +1 -1
- package/dist/tools/kb-query.d.ts +2 -2
- package/dist/tools/kb-reconcile-prefilter.d.ts.map +1 -1
- package/dist/tools/kb-reconcile-prefilter.js +3 -1
- package/dist/tools/kb-reconcile-prefilter.js.map +1 -1
- package/dist/tools/kb-resolve-contradiction.d.ts.map +1 -1
- package/dist/tools/kb-resolve-contradiction.js +3 -1
- package/dist/tools/kb-resolve-contradiction.js.map +1 -1
- package/dist/tools/kb-setup.d.ts.map +1 -1
- package/dist/tools/kb-setup.js +125 -20
- package/dist/tools/kb-setup.js.map +1 -1
- package/dist/tools/kb-stats.d.ts.map +1 -1
- package/dist/tools/kb-stats.js +64 -43
- package/dist/tools/kb-stats.js.map +1 -1
- package/dist/tools/list-members.d.ts +2 -2
- package/dist/tools/list-members.d.ts.map +1 -1
- package/dist/tools/list-members.js +1 -0
- package/dist/tools/list-members.js.map +1 -1
- package/dist/tools/member-detail.d.ts +2 -2
- package/dist/tools/member-detail.d.ts.map +1 -1
- package/dist/tools/member-detail.js +17 -0
- package/dist/tools/member-detail.js.map +1 -1
- package/dist/tools/member-reservation.d.ts +58 -1
- package/dist/tools/member-reservation.d.ts.map +1 -1
- package/dist/tools/member-reservation.js +61 -20
- package/dist/tools/member-reservation.js.map +1 -1
- package/dist/tools/provision-auth.d.ts +80 -1
- package/dist/tools/provision-auth.d.ts.map +1 -1
- package/dist/tools/provision-auth.js +142 -54
- package/dist/tools/provision-auth.js.map +1 -1
- package/dist/tools/provision-vcs-auth.d.ts +91 -1
- package/dist/tools/provision-vcs-auth.d.ts.map +1 -1
- package/dist/tools/provision-vcs-auth.js +209 -49
- package/dist/tools/provision-vcs-auth.js.map +1 -1
- package/dist/tools/register-member.d.ts +12 -9
- package/dist/tools/register-member.d.ts.map +1 -1
- package/dist/tools/register-member.js +99 -12
- package/dist/tools/register-member.js.map +1 -1
- package/dist/tools/remove-member.d.ts.map +1 -1
- package/dist/tools/remove-member.js +27 -2
- package/dist/tools/remove-member.js.map +1 -1
- package/dist/tools/send-email.d.ts +2 -2
- package/dist/tools/setup-git-app.d.ts.map +1 -1
- package/dist/tools/setup-git-app.js +11 -7
- package/dist/tools/setup-git-app.js.map +1 -1
- package/dist/tools/shutdown-server.d.ts +2 -0
- package/dist/tools/shutdown-server.d.ts.map +1 -1
- package/dist/tools/shutdown-server.js +23 -1
- package/dist/tools/shutdown-server.js.map +1 -1
- package/dist/tools/update-member.d.ts +12 -9
- package/dist/tools/update-member.d.ts.map +1 -1
- package/dist/tools/update-member.js +43 -11
- package/dist/tools/update-member.js.map +1 -1
- package/dist/tools/vcs-credential-exec.d.ts +127 -0
- package/dist/tools/vcs-credential-exec.d.ts.map +1 -0
- package/dist/tools/vcs-credential-exec.js +197 -0
- package/dist/tools/vcs-credential-exec.js.map +1 -0
- package/dist/types.d.ts +13 -0
- package/dist/types.d.ts.map +1 -1
- package/dist/utils/log-helpers.d.ts.map +1 -1
- package/dist/utils/log-helpers.js +2 -2
- package/dist/utils/log-helpers.js.map +1 -1
- package/dist/utils/prompt-errors.d.ts.map +1 -1
- package/dist/utils/prompt-errors.js +7 -1
- package/dist/utils/prompt-errors.js.map +1 -1
- package/dist/utils/shell-escape.d.ts +16 -0
- package/dist/utils/shell-escape.d.ts.map +1 -1
- package/dist/utils/shell-escape.js +22 -2
- package/dist/utils/shell-escape.js.map +1 -1
- package/dist/utils/vcs-provider-detect.d.ts +65 -0
- package/dist/utils/vcs-provider-detect.d.ts.map +1 -0
- package/dist/utils/vcs-provider-detect.js +109 -0
- package/dist/utils/vcs-provider-detect.js.map +1 -0
- package/dist/utils/workspace-trust.d.ts +20 -17
- package/dist/utils/workspace-trust.d.ts.map +1 -1
- package/dist/utils/workspace-trust.js +87 -1
- package/dist/utils/workspace-trust.js.map +1 -1
- package/dist/workflows/auto-sprint.js +176 -15
- package/package.json +6 -4
- package/packages/apra-fleet-client/src/client/api.mjs +319 -17
- package/packages/apra-fleet-client/src/client/errors.mjs +6 -2
- package/packages/apra-fleet-se/apra-pm/agents/_shared/GRAPH-SEMANTICS.md +10 -2
- package/packages/apra-fleet-se/apra-pm/agents/backlog-groomer.md +25 -0
- package/packages/apra-fleet-se/apra-pm/agents/ci-watcher.md +11 -0
- package/packages/apra-fleet-se/apra-pm/agents/deployer.md +37 -2
- package/packages/apra-fleet-se/apra-pm/agents/doer.md +36 -12
- package/packages/apra-fleet-se/apra-pm/agents/harvester.md +11 -0
- package/packages/apra-fleet-se/apra-pm/agents/integ-test-runner.md +11 -0
- package/packages/apra-fleet-se/apra-pm/agents/kb-reconciler.md +9 -0
- package/packages/apra-fleet-se/apra-pm/agents/plan-reviewer.md +110 -22
- package/packages/apra-fleet-se/apra-pm/agents/planner.md +125 -37
- package/packages/apra-fleet-se/apra-pm/agents/regression-test-runner.md +11 -0
- package/packages/apra-fleet-se/apra-pm/agents/reviewer.md +13 -2
- package/packages/apra-fleet-se/apra-pm/agents/schemas/plan-reviewer-output.json +37 -0
- package/packages/apra-fleet-se/apra-pm/skills/pm/SKILL.md +1 -1
- package/packages/apra-fleet-se/apra-pm/skills/pm/cost.md +0 -3
- package/packages/apra-fleet-se/apra-pm/skills/pm/fleet-addendum.md +2 -2
- package/packages/apra-fleet-se/bin/cli.mjs +149 -1
- package/packages/apra-fleet-se/bin/serve.mjs +78 -4
- package/packages/apra-fleet-se/fleet-sprint/abort.mjs +481 -0
- package/packages/apra-fleet-se/fleet-sprint/beads-children.mjs +491 -0
- package/packages/apra-fleet-se/fleet-sprint/beads-identity-check.mjs +282 -0
- package/packages/apra-fleet-se/fleet-sprint/beads-identity.mjs +208 -0
- package/packages/apra-fleet-se/fleet-sprint/beads-scope.mjs +638 -0
- package/packages/apra-fleet-se/fleet-sprint/beads-transitions.mjs +301 -0
- package/packages/apra-fleet-se/fleet-sprint/branch-ensure.mjs +96 -0
- package/packages/apra-fleet-se/fleet-sprint/conflict-ladder.mjs +8 -0
- package/packages/apra-fleet-se/fleet-sprint/contracts.mjs +61 -14
- package/packages/apra-fleet-se/fleet-sprint/coordination.mjs +516 -0
- package/packages/apra-fleet-se/fleet-sprint/dispatch-failure.mjs +184 -0
- package/packages/apra-fleet-se/fleet-sprint/dispatch-role.mjs +1121 -0
- package/packages/apra-fleet-se/fleet-sprint/dispatch-safety-guard.mjs +216 -4
- package/packages/apra-fleet-se/fleet-sprint/docs/README.md +1 -1
- package/packages/apra-fleet-se/fleet-sprint/docs/dolt-sync-redesign.md +276 -0
- package/packages/apra-fleet-se/fleet-sprint/docs/escalate-to-llm-design.md +1 -1
- package/packages/apra-fleet-se/fleet-sprint/docs/run-sprint-cycle-prelude-decision.md +215 -0
- package/packages/apra-fleet-se/fleet-sprint/dolt-literal-guard.mjs +56 -1
- package/packages/apra-fleet-se/fleet-sprint/dolt-sync.mjs +996 -32
- package/packages/apra-fleet-se/fleet-sprint/errors.mjs +267 -0
- package/packages/apra-fleet-se/fleet-sprint/explicit-id-create-guard.mjs +501 -0
- package/packages/apra-fleet-se/fleet-sprint/fatal-diagnostics.mjs +134 -0
- package/packages/apra-fleet-se/fleet-sprint/full-db-fetch-guard.mjs +102 -0
- package/packages/apra-fleet-se/fleet-sprint/git-sync.mjs +556 -0
- package/packages/apra-fleet-se/fleet-sprint/git-topology.mjs +495 -0
- package/packages/apra-fleet-se/fleet-sprint/guarded-modules.mjs +564 -0
- package/packages/apra-fleet-se/fleet-sprint/inline-ladder-guard.mjs +216 -0
- package/packages/apra-fleet-se/fleet-sprint/kb.mjs +728 -0
- package/packages/apra-fleet-se/fleet-sprint/mcp-result.mjs +75 -0
- package/packages/apra-fleet-se/fleet-sprint/member-provisioning.mjs +252 -0
- package/packages/apra-fleet-se/fleet-sprint/member-sync.mjs +581 -0
- package/packages/apra-fleet-se/fleet-sprint/member-target.mjs +93 -0
- package/packages/apra-fleet-se/fleet-sprint/newtask-text.mjs +266 -0
- package/packages/apra-fleet-se/fleet-sprint/parent-notes-staleness.mjs +201 -0
- package/packages/apra-fleet-se/fleet-sprint/phases/deploy.mjs +136 -0
- package/packages/apra-fleet-se/fleet-sprint/phases/develop.mjs +822 -0
- package/packages/apra-fleet-se/fleet-sprint/phases/ensure-sprint-branch.mjs +294 -0
- package/packages/apra-fleet-se/fleet-sprint/phases/final-review.mjs +403 -0
- package/packages/apra-fleet-se/fleet-sprint/phases/harvest.mjs +180 -0
- package/packages/apra-fleet-se/fleet-sprint/phases/integ-test.mjs +358 -0
- package/packages/apra-fleet-se/fleet-sprint/phases/plan.mjs +390 -0
- package/packages/apra-fleet-se/fleet-sprint/phases/publish-pr.mjs +323 -0
- package/packages/apra-fleet-se/fleet-sprint/phases/re-review.mjs +232 -0
- package/packages/apra-fleet-se/fleet-sprint/phases/regression-test.mjs +174 -0
- package/packages/apra-fleet-se/fleet-sprint/phases/replan.mjs +197 -0
- package/packages/apra-fleet-se/fleet-sprint/phases/review.mjs +305 -0
- package/packages/apra-fleet-se/fleet-sprint/prompts.mjs +559 -0
- package/packages/apra-fleet-se/fleet-sprint/role-policies.mjs +1416 -0
- package/packages/apra-fleet-se/fleet-sprint/round-session.mjs +100 -0
- package/packages/apra-fleet-se/fleet-sprint/runner.js +1467 -9798
- package/packages/apra-fleet-se/fleet-sprint/se-posix.mjs +17 -12
- package/packages/apra-fleet-se/fleet-sprint/shell-command-guard.mjs +304 -0
- package/packages/apra-fleet-se/fleet-sprint/skills/fleet-sprint-cli/SKILL.md +18 -0
- package/packages/apra-fleet-se/fleet-sprint/skills/fleet-supervisor/SKILL.md +21 -2
- package/packages/apra-fleet-se/fleet-sprint/sprint-args.mjs +454 -0
- package/packages/apra-fleet-se/fleet-sprint/sprint-report.mjs +239 -0
- package/packages/apra-fleet-se/fleet-sprint/sprint-state.mjs +163 -0
- package/packages/apra-fleet-se/fleet-sprint/unbracketed-push-guard.mjs +234 -0
- package/packages/apra-fleet-se/fleet-sprint/usage-limit-controller.mjs +211 -0
- package/packages/apra-fleet-se/fleet-sprint/vcs-auth.mjs +1279 -0
- package/packages/apra-fleet-se/fleet-sprint/vcs-module.mjs +153 -19
- package/packages/apra-fleet-se/fleet-sprint/vcs-providers/azure-devops.mjs +71 -21
- package/packages/apra-fleet-se/fleet-sprint/vcs-providers/bitbucket.mjs +28 -0
- package/packages/apra-fleet-se/fleet-sprint/vcs-providers/github.mjs +131 -0
- package/packages/apra-fleet-se/fleet-sprint/vcs-providers/index.mjs +140 -16
- package/packages/apra-fleet-se/fleet-sprint/vcs-providers/shell-helpers.mjs +1 -0
- package/packages/apra-fleet-se/fleet-sprint/viewer-extensions.mjs +77 -2
- package/packages/apra-fleet-se/fleet-sprint/worklists.mjs +559 -0
- package/packages/apra-fleet-se/package.json +1 -1
- package/packages/apra-fleet-workflow/src/viewer/index.mjs +17 -6
- package/packages/apra-fleet-workflow/src/workflow/index.mjs +49 -7
- package/skills/fleet/SKILL.md +8 -8
- package/skills/fleet/auth-azdevops.md +14 -3
- package/skills/fleet/auth-bitbucket.md +2 -2
- package/skills/fleet/auth-github.md +5 -4
- package/skills/fleet/onboarding.md +3 -3
- package/skills/fleet/permissions.md +5 -1
- package/skills/fleet/profiles/node.json +1 -1
- package/skills/fleet/troubleshooting.md +3 -3
- package/version.json +1 -1
|
@@ -1,5 +1,6 @@
|
|
|
1
1
|
import fs from 'fs';
|
|
2
2
|
import path from 'path';
|
|
3
|
+
import { guardedModulePaths } from './guarded-modules.mjs';
|
|
3
4
|
|
|
4
5
|
// =============================================================================
|
|
5
6
|
// apra-fleet-eft.3.3 -- dispatch-safety guard checker, extracted so it is
|
|
@@ -19,6 +20,16 @@ import path from 'path';
|
|
|
19
20
|
// Invariant under test (unchanged from eft.3.1): EVERY `command(` / `agent(`
|
|
20
21
|
// call site in a scanned source file must supply an explicit `member_name`
|
|
21
22
|
// (or `member_id`) in its options object.
|
|
23
|
+
//
|
|
24
|
+
// MODULE-LIST GENERALIZATION: this guard is no longer pointed at a single
|
|
25
|
+
// hard-coded file. checkModules() below reads the SHARED guarded-module list
|
|
26
|
+
// (./guarded-modules.mjs) -- the one place a newly extracted fleet-sprint
|
|
27
|
+
// module is registered -- so a dispatch that moves out of runner.js into a
|
|
28
|
+
// new module stays covered instead of silently falling out of scan scope.
|
|
29
|
+
// checkPath() remains exported and behaves exactly as before; it is still the
|
|
30
|
+
// right entry point for scanning one specific file (a fixture, or a module
|
|
31
|
+
// like dolt-sync.mjs that is deliberately not on the shared list). This
|
|
32
|
+
// module is the reference implementation the sibling guards follow.
|
|
22
33
|
// =============================================================================
|
|
23
34
|
|
|
24
35
|
/**
|
|
@@ -63,18 +74,177 @@ export function skipStringLiteral(src, start, quoteChar) {
|
|
|
63
74
|
return i;
|
|
64
75
|
}
|
|
65
76
|
|
|
77
|
+
// A `/` that opens a REGEX literal (as opposed to a division/`/=` operator)
|
|
78
|
+
// is only ever preceded -- skipping whitespace -- by one of these punctuation
|
|
79
|
+
// characters, by the start of the file, or by one of the listed keywords.
|
|
80
|
+
// This is the same heuristic real tokenizers use to disambiguate the two: a
|
|
81
|
+
// division operator always follows a value (an identifier, number, `)`, `]`,
|
|
82
|
+
// `}`, or a closing quote), never one of these. Good enough for source text
|
|
83
|
+
// that is valid, already-linted JS (this repo's own modules), which is the
|
|
84
|
+
// only input maskComments() is ever run against.
|
|
85
|
+
const REGEX_PRECEDING_PUNCT_CHARS = new Set('([{,;:=!&|?+-*%^~<>'.split(''));
|
|
86
|
+
const REGEX_PRECEDING_KEYWORDS = new Set([
|
|
87
|
+
'return', 'typeof', 'instanceof', 'in', 'of', 'new', 'delete', 'void',
|
|
88
|
+
'yield', 'throw', 'case', 'do', 'else', 'await',
|
|
89
|
+
]);
|
|
90
|
+
|
|
91
|
+
/**
|
|
92
|
+
* True when a `/` at position `idx` in `text` opens a regex literal rather
|
|
93
|
+
* than a division or `/=` operator. Scans backward from `idx` (skipping
|
|
94
|
+
* whitespace) to the last significant character/word and checks it against
|
|
95
|
+
* REGEX_PRECEDING_PUNCT_CHARS/REGEX_PRECEDING_KEYWORDS above. `text` must
|
|
96
|
+
* already have comments masked to whitespace at every position before `idx`
|
|
97
|
+
* (both maskComments()'s own in-progress `out` buffer and its finished
|
|
98
|
+
* return value satisfy this), so a `/` right after where a comment used to
|
|
99
|
+
* be correctly sees whatever preceded the comment, not the comment itself.
|
|
100
|
+
* EXPORTED so any OTHER walk over an already-masked copy of source (e.g. a
|
|
101
|
+
* depth-count loop that must not miscount a quote or bracket embedded in a
|
|
102
|
+
* regex body) can reuse the exact same regex-vs-division heuristic instead
|
|
103
|
+
* of re-deriving it -- see extractBalancedCall() below and
|
|
104
|
+
* unbracketed-push-guard.mjs's findFunctionBodyRange() for two consumers
|
|
105
|
+
* that got this wrong before apra-fleet-btj9.10.
|
|
106
|
+
*/
|
|
107
|
+
export function canStartRegex(text, idx) {
|
|
108
|
+
let j = idx - 1;
|
|
109
|
+
while (j >= 0 && /\s/.test(text[j])) j--;
|
|
110
|
+
if (j < 0) return true; // start of file
|
|
111
|
+
if (REGEX_PRECEDING_PUNCT_CHARS.has(text[j])) return true;
|
|
112
|
+
let k = j;
|
|
113
|
+
while (k >= 0 && /[A-Za-z0-9_$]/.test(text[k])) k--;
|
|
114
|
+
const word = text.slice(k + 1, j + 1);
|
|
115
|
+
return REGEX_PRECEDING_KEYWORDS.has(word);
|
|
116
|
+
}
|
|
117
|
+
|
|
118
|
+
/**
|
|
119
|
+
* Returns the index of the closing `/` of a regex literal that opens at
|
|
120
|
+
* `start`, or -1 if none is found before end-of-line/end-of-file (a regex
|
|
121
|
+
* literal can never span a raw newline in valid JS, so hitting one means the
|
|
122
|
+
* leading `/` was misclassified as a regex start -- the caller falls back to
|
|
123
|
+
* treating it as an ordinary character). Tracks bracket-expression state
|
|
124
|
+
* (`[...]`) because a `/` inside a character class does not close the regex
|
|
125
|
+
* (this is exactly what SAFE_TEXT_RE = /^[A-Za-z0-9 .,:;!?()'_/+[\]-]+$/ in
|
|
126
|
+
* newtask-text.mjs relies on), and skips backslash-escaped characters
|
|
127
|
+
* (inside or outside the class) the same way skipStringLiteral() does.
|
|
128
|
+
* EXPORTED for the same reason as canStartRegex() above.
|
|
129
|
+
*/
|
|
130
|
+
export function skipRegexLiteral(src, start) {
|
|
131
|
+
let inClass = false;
|
|
132
|
+
for (let i = start + 1; i < src.length; i++) {
|
|
133
|
+
const ch = src[i];
|
|
134
|
+
if (ch === '\\') { i++; continue; }
|
|
135
|
+
if (ch === '\n') return -1;
|
|
136
|
+
if (inClass) {
|
|
137
|
+
if (ch === ']') inClass = false;
|
|
138
|
+
continue;
|
|
139
|
+
}
|
|
140
|
+
if (ch === '[') { inClass = true; continue; }
|
|
141
|
+
if (ch === '/') return i;
|
|
142
|
+
}
|
|
143
|
+
return -1;
|
|
144
|
+
}
|
|
145
|
+
|
|
146
|
+
/**
|
|
147
|
+
* Replaces every comment's characters with spaces (newlines preserved), so
|
|
148
|
+
* the result has the SAME length and line numbering as `src` but no comment
|
|
149
|
+
* text -- mirrors test/helpers/dispatch-pin-scanner.mjs's maskComments()
|
|
150
|
+
* (that file's stripComments() helper is NOT length-preserving, which would
|
|
151
|
+
* corrupt the positional index math extractBalancedCall depends on).
|
|
152
|
+
*
|
|
153
|
+
* WHY THIS EXISTS (apra-fleet-3swo.34): extractBalancedCall()'s depth walk
|
|
154
|
+
* skips over string literals but, before this fix, did not skip comments.
|
|
155
|
+
* runner.js's prose comments are full of apostrophes (e.g. "the streak's
|
|
156
|
+
* scope"); an unmasked walk reads that apostrophe as an opening quote and
|
|
157
|
+
* swallows everything -- including real closing parens -- until the next
|
|
158
|
+
* apostrophe, so a call site's balanced range can silently run away to the
|
|
159
|
+
* end of the file. Masking comments out before walking parens/quotes is what
|
|
160
|
+
* makes the balanced range trustworthy; the returned callText is still
|
|
161
|
+
* sliced from the ORIGINAL (unmasked) src so callers keep seeing real
|
|
162
|
+
* comment text, just with correct boundaries.
|
|
163
|
+
*
|
|
164
|
+
* REGEX LITERALS (apra-fleet-btj9.10): a regex whose body contains a quote or
|
|
165
|
+
* apostrophe -- e.g. newtask-text.mjs's
|
|
166
|
+
* `SAFE_TEXT_RE = /^[A-Za-z0-9 .,:;!?()'_/+[\]-]+$/` or branch-ensure.mjs's
|
|
167
|
+
* `/couldn't find remote ref/i` -- used to open the SAME kind of phantom
|
|
168
|
+
* string the comment-blindness fix above closed for apostrophes in prose: the
|
|
169
|
+
* string branch below would treat that quote as opening a real string and
|
|
170
|
+
* copy everything up to the next matching quote through UNMASKED, including
|
|
171
|
+
* any real comment in between (whose stray parens could then desync a later
|
|
172
|
+
* extractBalancedCall() depth walk). Regex literals are recognized (via
|
|
173
|
+
* canStartRegex()/skipRegexLiteral() above) and copied through verbatim --
|
|
174
|
+
* like a string, not stripped -- BEFORE the string branch gets a chance to
|
|
175
|
+
* misread the quote inside one.
|
|
176
|
+
*
|
|
177
|
+
* @param {string} src
|
|
178
|
+
* @returns {string}
|
|
179
|
+
*/
|
|
180
|
+
export function maskComments(src) {
|
|
181
|
+
let out = '';
|
|
182
|
+
for (let i = 0; i < src.length; i++) {
|
|
183
|
+
const ch = src[i];
|
|
184
|
+
if (ch === '/' && src[i + 1] === '/') {
|
|
185
|
+
while (i < src.length && src[i] !== '\n') { out += ' '; i++; }
|
|
186
|
+
out += '\n';
|
|
187
|
+
continue;
|
|
188
|
+
}
|
|
189
|
+
if (ch === '/' && src[i + 1] === '*') {
|
|
190
|
+
const end = src.indexOf('*/', i + 2);
|
|
191
|
+
const stop = end < 0 ? src.length - 1 : end + 1;
|
|
192
|
+
for (; i <= stop; i++) out += src[i] === '\n' ? '\n' : ' ';
|
|
193
|
+
i--;
|
|
194
|
+
continue;
|
|
195
|
+
}
|
|
196
|
+
if (ch === '/' && canStartRegex(out, out.length)) {
|
|
197
|
+
const end = skipRegexLiteral(src, i);
|
|
198
|
+
if (end !== -1) {
|
|
199
|
+
let flagEnd = end;
|
|
200
|
+
while (flagEnd + 1 < src.length && /[a-zA-Z]/.test(src[flagEnd + 1])) flagEnd++;
|
|
201
|
+
out += src.slice(i, flagEnd + 1);
|
|
202
|
+
i = flagEnd;
|
|
203
|
+
continue;
|
|
204
|
+
}
|
|
205
|
+
// Not actually a regex (no closing `/` before end of line) --
|
|
206
|
+
// fall through and let the default branch below copy the `/`
|
|
207
|
+
// as an ordinary character.
|
|
208
|
+
}
|
|
209
|
+
if (ch === '"' || ch === "'" || ch === '`') {
|
|
210
|
+
const end = skipStringLiteral(src, i, ch);
|
|
211
|
+
out += src.slice(i, end + 1);
|
|
212
|
+
i = end;
|
|
213
|
+
continue;
|
|
214
|
+
}
|
|
215
|
+
out += ch;
|
|
216
|
+
}
|
|
217
|
+
return out;
|
|
218
|
+
}
|
|
219
|
+
|
|
66
220
|
/**
|
|
67
221
|
* Given the index of an opening '(' in `src`, returns the full call-site
|
|
68
222
|
* text from that '(' through its matching ')', tracking paren depth and
|
|
69
223
|
* skipping over string/template-literal contents (so parens embedded in
|
|
70
224
|
* string/template content, e.g. `bd show ${ids.join(' ')}`, never disturb
|
|
71
|
-
* the depth count).
|
|
225
|
+
* the depth count) AND over comment spans (so an apostrophe in prose, e.g.
|
|
226
|
+
* "the streak's scope", is never misread as opening a string -- see
|
|
227
|
+
* maskComments() above). The depth/quote walk runs over a comment-masked
|
|
228
|
+
* copy of `src`, but the returned text is sliced from the ORIGINAL `src` so
|
|
229
|
+
* real comment content is preserved in the output.
|
|
230
|
+
*
|
|
231
|
+
* ALSO skips over regex literals in this SAME walk (apra-fleet-btj9.10):
|
|
232
|
+
* maskComments() copies a regex literal's body through verbatim (like a
|
|
233
|
+
* string's), so a quote/apostrophe inside one (e.g. a call argument written
|
|
234
|
+
* `/it's ok/`) is still sitting right there in `masked`. Without this,
|
|
235
|
+
* THIS walk's own quote branch below would re-discover that apostrophe and
|
|
236
|
+
* misread it as opening a real string, potentially running the depth walk
|
|
237
|
+
* away past the call's real closing paren. Uses the same
|
|
238
|
+
* canStartRegex()/skipRegexLiteral() maskComments() itself uses, so a
|
|
239
|
+
* `/` is only ever treated as a regex open here in the exact cases it would
|
|
240
|
+
* have been treated as one during masking.
|
|
72
241
|
*/
|
|
73
242
|
export function extractBalancedCall(src, openParenIdx) {
|
|
243
|
+
const masked = maskComments(src);
|
|
74
244
|
let depth = 0;
|
|
75
245
|
let i = openParenIdx;
|
|
76
|
-
for (; i <
|
|
77
|
-
const ch =
|
|
246
|
+
for (; i < masked.length; i++) {
|
|
247
|
+
const ch = masked[i];
|
|
78
248
|
if (ch === '(') {
|
|
79
249
|
depth++;
|
|
80
250
|
} else if (ch === ')') {
|
|
@@ -82,8 +252,11 @@ export function extractBalancedCall(src, openParenIdx) {
|
|
|
82
252
|
if (depth === 0) {
|
|
83
253
|
return src.slice(openParenIdx, i + 1);
|
|
84
254
|
}
|
|
255
|
+
} else if (ch === '/' && canStartRegex(masked, i)) {
|
|
256
|
+
const end = skipRegexLiteral(masked, i);
|
|
257
|
+
if (end !== -1) i = end;
|
|
85
258
|
} else if (ch === '"' || ch === "'" || ch === '`') {
|
|
86
|
-
i = skipStringLiteral(
|
|
259
|
+
i = skipStringLiteral(masked, i, ch);
|
|
87
260
|
}
|
|
88
261
|
}
|
|
89
262
|
// Unbalanced -- should never happen against real, syntactically-valid
|
|
@@ -175,3 +348,42 @@ export function checkPath(filePath) {
|
|
|
175
348
|
const violations = findViolations(sites, fileLabel);
|
|
176
349
|
return { sites, violations };
|
|
177
350
|
}
|
|
351
|
+
|
|
352
|
+
/**
|
|
353
|
+
* Aggregate entry point: scans EVERY module in the shared guarded-module list
|
|
354
|
+
* (fleet-sprint/guarded-modules.mjs -- runner.js today, plus whatever is
|
|
355
|
+
* extracted from it) and returns the union of their call sites and
|
|
356
|
+
* violations. This is the reference implementation the other mechanical
|
|
357
|
+
* guards in this directory follow.
|
|
358
|
+
*
|
|
359
|
+
* Each violation string already carries the basename of the file it came
|
|
360
|
+
* from (findViolations' `fileLabel`), so an aggregate run over several
|
|
361
|
+
* modules attributes every finding to the correct module rather than to a
|
|
362
|
+
* single hard-coded runner.js label. `sitesByFile` gives per-module call-site
|
|
363
|
+
* counts for baseline assertions that need them.
|
|
364
|
+
*
|
|
365
|
+
* `paths` defaults to the shared list; callers pass
|
|
366
|
+
* guardedModulePaths([fixture]) to prove the list -- not a hard-coded name --
|
|
367
|
+
* is what the guard actually reads.
|
|
368
|
+
*
|
|
369
|
+
* @param {string[]} [paths]
|
|
370
|
+
* @returns {{ sites: object[], violations: string[], files: string[], sitesByFile: Record<string, object[]> }}
|
|
371
|
+
*/
|
|
372
|
+
export function checkModules(paths = guardedModulePaths()) {
|
|
373
|
+
if (!Array.isArray(paths)) {
|
|
374
|
+
throw new TypeError('checkModules(paths): paths must be an array of file paths');
|
|
375
|
+
}
|
|
376
|
+
const sites = [];
|
|
377
|
+
const violations = [];
|
|
378
|
+
const files = [];
|
|
379
|
+
const sitesByFile = {};
|
|
380
|
+
for (const p of paths) {
|
|
381
|
+
const file = path.basename(p);
|
|
382
|
+
const result = checkPath(p);
|
|
383
|
+
files.push(file);
|
|
384
|
+
sitesByFile[file] = result.sites;
|
|
385
|
+
sites.push(...result.sites.map((s) => ({ ...s, file })));
|
|
386
|
+
violations.push(...result.violations);
|
|
387
|
+
}
|
|
388
|
+
return { sites, violations, files, sitesByFile };
|
|
389
|
+
}
|
|
@@ -90,7 +90,7 @@ Watch progress there rather than tailing raw stdout.
|
|
|
90
90
|
| `--role-map <json\|@file>` | | no | -- | JSON object mapping role -> member[], e.g. `'{"doer":["m1","m2"]}'`. Inline JSON or `@path/to/file.json`. |
|
|
91
91
|
| `--viewer-port <port>` | | no | `8080` | Port for the local dashboard viewer. |
|
|
92
92
|
| `--budget <usd>` | | no | unlimited | USD ceiling for this run's total estimated spend. |
|
|
93
|
-
| `--dispatch-timeout-s <s>` | | no | `9000` | Per-dispatch time budget in seconds
|
|
93
|
+
| `--dispatch-timeout-s <s>` | | no | `9000` | Per-dispatch total time budget in seconds -- the hard ceiling on every agent dispatch (integ-test dispatch ceiling is 2x, regression-test ceiling is 3x). Minimum 60. The inactivity/stall threshold used to detect a hung dispatch is derived from this value but capped separately at 30 minutes, so raising this budget for a large sprint does not also grant a stalled dispatch a correspondingly longer grace period before it is killed. Lower it for small sprints so a hung dispatch costs minutes, not 2.5 hours. |
|
|
94
94
|
| `--sync` | | no | off | Synced topology mode (orchestrator-bracketed git sync): members may sit on differing HEADs but must share the same origin URL and pass a `bd dolt pull` probe. Omitted uses shared-workspace mode (all members on the same HEAD). |
|
|
95
95
|
| `--help` | `-h` | | | Show the engine's help. |
|
|
96
96
|
|
|
@@ -1193,6 +1193,57 @@ For the target member M, starting from both clones at the same sandbox head:
|
|
|
1193
1193
|
declared verified. It is intentionally NOT in any default CI pass, and no
|
|
1194
1194
|
CI configuration should be added that pretends otherwise.
|
|
1195
1195
|
|
|
1196
|
+
### 6.7 Re-affirmed CI-exclusion decision (apra-fleet-j918.5.3, current pass)
|
|
1197
|
+
|
|
1198
|
+
**Decided 2026-09-15, against tree `f4eda6e08e53412c5cffe5bae25d686280987cbd`
|
|
1199
|
+
(branch `chore/j918-test-suite-cleanup`).** A test-suite audit flagged that
|
|
1200
|
+
the merge-conflict recovery path (`settleDoltConflicts()`,
|
|
1201
|
+
`fleet-sprint/dolt-settle.mjs`) is exercised in `npm test` ONLY by
|
|
1202
|
+
`test/dolt-settle.test.mjs`'s hand-written SQL-string simulator -- a scripted
|
|
1203
|
+
`command()` mock that regex-matches emitted SQL text (`CALL DOLT_MERGE`,
|
|
1204
|
+
`SELECT * FROM dolt_conflicts`, `CALL DOLT_CONFLICTS_RESOLVE`, ...; see e.g.
|
|
1205
|
+
`test/dolt-settle.test.mjs:389-423`) rather than running a real `dolt`
|
|
1206
|
+
process. This 6.2/6.6 exclusion was decided in 2026-08 for a different
|
|
1207
|
+
reason (no generic CI runner has a live fleet server or SSH-reachable
|
|
1208
|
+
members) and never separately weighed against the fact that CI's only
|
|
1209
|
+
coverage of the real recovery path is that simulator. Re-examining now
|
|
1210
|
+
because the failure mode this path guards -- a Dolt-CLI/SQL-dialect
|
|
1211
|
+
regression silently corrupting or losing rows in the SHARED beads DB during
|
|
1212
|
+
a real merge -- is a data-loss risk, and an inherited justification for a
|
|
1213
|
+
different concern is not evidence this specific gap is acceptable.
|
|
1214
|
+
|
|
1215
|
+
**The header's exclusion claim was re-verified against the current file
|
|
1216
|
+
before writing this section** (`scripts/dolt-settle-integration.mjs:24-27`):
|
|
1217
|
+
it still requires a live `apra-fleet` server (`ApraFleet`/`McpClient` over
|
|
1218
|
+
`resolveFleetServerConnection`) plus SSH-reachable members
|
|
1219
|
+
(`createMemberReservationClient`, real `execute_command` dispatch) -- nothing
|
|
1220
|
+
about that dependency has changed since Part 6.2, so **this script itself
|
|
1221
|
+
stays excluded from `npm test`/`test:unit`/`test:slow`, unchanged.** Its
|
|
1222
|
+
live-fleet/SSH requirement is a hard blocker for any generic CI runner
|
|
1223
|
+
regardless of how cheap `dolt` itself is to run.
|
|
1224
|
+
|
|
1225
|
+
That said, the data-loss coverage gap is real and worth closing separately,
|
|
1226
|
+
because it does not actually require this script's live-fleet/SSH machinery:
|
|
1227
|
+
`dolt` is a single, statically-linked binary, already version-pinned
|
|
1228
|
+
(`DOLT_VERSION`, Part 5.1), and `settleDoltConflicts()` only needs an
|
|
1229
|
+
injected `command()` -- it never requires the fleet/SSH transport itself.
|
|
1230
|
+
A much narrower job -- two local `bd` clones on the CI runner, a local
|
|
1231
|
+
merge conflict manufactured directly (no fleet server, no member
|
|
1232
|
+
reservation, no SSH), and `settleDoltConflicts()` wired to a `command()`
|
|
1233
|
+
that shells out to the pinned local `dolt` binary -- is plausible without
|
|
1234
|
+
reproducing any of the live-fleet dependency that justifies excluding
|
|
1235
|
+
*this* script. Weighed against that: added CI runtime, and the blast
|
|
1236
|
+
radius of any CI job that manipulates real Dolt merge state (mitigated by
|
|
1237
|
+
running entirely in a disposable local temp dir, same isolation principle
|
|
1238
|
+
as Part 6.4, never touching a shared remote).
|
|
1239
|
+
|
|
1240
|
+
**Decision: add that narrower, local-only, live-`dolt` test as a follow-up,
|
|
1241
|
+
filed as apra-fleet-j918.9 (linked from this doc; not implemented by this
|
|
1242
|
+
task).** `scripts/dolt-settle-integration.mjs` remains excluded from CI as
|
|
1243
|
+
Part 6.2/6.6 already state -- this section adds the current-pass
|
|
1244
|
+
justification and the follow-up, it does not change the exclusion itself.
|
|
1245
|
+
No CI configuration is added by this task.
|
|
1246
|
+
|
|
1196
1247
|
---
|
|
1197
1248
|
|
|
1198
1249
|
## Part 7 -- persistent vs. ephemeral `dolt sql-server`, answered precisely
|
|
@@ -1485,3 +1536,228 @@ so the list below reads as history rather than as a pending action:
|
|
|
1485
1536
|
integrity bug (a filed bead reference vanished), but not a sync/conflict
|
|
1486
1537
|
issue and not something settle() affects. Left open, unrelated.
|
|
1487
1538
|
|
|
1539
|
+
## Part 9 -- sync COST: fewer and cheaper syncs per sprint
|
|
1540
|
+
|
|
1541
|
+
Parts 1-8 are about sync CORRECTNESS (what happens when a sync conflicts).
|
|
1542
|
+
This part is about sync COST, from a separate read-only review of where a
|
|
1543
|
+
fleet-sprint actually spends its `bd dolt pull` / `bd dolt push` minutes. The
|
|
1544
|
+
measured shape on the reference machine: about 20 D-pull/D-push pairs per
|
|
1545
|
+
sprint, each bracketed by 1-2 uncached `bd config get sync.remote` probes
|
|
1546
|
+
(0.6s each), against a 548 MB embedded Dolt store whose full open-and-walk
|
|
1547
|
+
cost is paid on every invocation regardless of how little changed.
|
|
1548
|
+
|
|
1549
|
+
Three changes landed in `dolt-sync.mjs` (a fourth, in the supervisor's
|
|
1550
|
+
pre-launch scope guard, is described in `src/supervisor/scope-overlap.mjs`).
|
|
1551
|
+
|
|
1552
|
+
### 9.1 The `sync.remote` probe is memoized per member
|
|
1553
|
+
|
|
1554
|
+
`isMemberSyncRemoteConfigured()` spawned a fresh `bd config get sync.remote
|
|
1555
|
+
--json` on every call -- the D-pull pre-gate, the D-push pre-gate, and
|
|
1556
|
+
`status()` -- which totals roughly 90-160 spawns per sprint for a value that
|
|
1557
|
+
never changes mid-run. A codebase search confirmed no production path in this
|
|
1558
|
+
package writes `sync.remote` during a live sprint; only test fixtures and
|
|
1559
|
+
one-time setup scripts do.
|
|
1560
|
+
|
|
1561
|
+
The result is now cached per member for the process lifetime (one runner
|
|
1562
|
+
process per sprint). Two properties matter:
|
|
1563
|
+
|
|
1564
|
+
- **Only a positively parsed answer is cached.** Every fail-safe path (the
|
|
1565
|
+
command threw, a `failSoft` error result, no output, unparseable output)
|
|
1566
|
+
still reports "configured" and is deliberately NOT cached, so one transient
|
|
1567
|
+
probe failure cannot pin the fail-safe answer for the rest of the run. The
|
|
1568
|
+
fail-CLOSED contract is unchanged.
|
|
1569
|
+
- **Explicit invalidation, not a TTL.** A TTL re-adds spawns for no real
|
|
1570
|
+
safety. Three HARD seams drop the memo, and every one of them drops the 9.3
|
|
1571
|
+
remote-tip fingerprint alongside it (the two are always forgotten
|
|
1572
|
+
together): `noteMemberCommand()`, called from the runner's central
|
|
1573
|
+
`command()` wrapper for any `bd config set` / `bd dolt remote` / `bd init` /
|
|
1574
|
+
`bd bootstrap` the orchestrator issues on that member; the auth self-heal
|
|
1575
|
+
firing inside `runDoltStep`; and `repair()`.
|
|
1576
|
+
- **The per-dispatch seam is SOFT (round 4).** A dispatched agent runs its
|
|
1577
|
+
`bd` commands in its own session on the member, never through the
|
|
1578
|
+
`command()` wrapper, so `noteMemberCommand()` cannot see them; the runner's
|
|
1579
|
+
central `agent()` wrapper calls `noteMemberDispatchCompleted()` the moment
|
|
1580
|
+
any dispatch settles. Round 3 made that an unconditional wipe of both
|
|
1581
|
+
memos. Combined with "a push never mints a fingerprint" (9.3), the
|
|
1582
|
+
fingerprint was then EMPTY at the start of every dispatch bracket -- the
|
|
1583
|
+
module's primary path -- so the primary path paid one `ls-remote` more
|
|
1584
|
+
than before the feature and skipped nothing, and the probe was re-spawned
|
|
1585
|
+
once per dispatch (golden mock-sprint transcript: 1 -> 14). The hazards
|
|
1586
|
+
the wipe guarded were checked against what bd actually does:
|
|
1587
|
+
`bd bootstrap` -- the one self-heal this repo's agent instructions
|
|
1588
|
+
prescribe -- is non-destructive (an existing DB is validated and reported,
|
|
1589
|
+
never replaced) and, where it creates a DB, CLONES it from `sync.remote`,
|
|
1590
|
+
so a surviving fingerprint that still equals the remote tip is still the
|
|
1591
|
+
truth; `bd init` refuses on an existing DB unless forced, and a forced
|
|
1592
|
+
re-init yields an unrelated history that no pull can fast-forward (its next
|
|
1593
|
+
push diverges into the terminals that already forget the tip and settle).
|
|
1594
|
+
The one event a surviving fingerprint would get WRONG is an agent-side
|
|
1595
|
+
`bd config set sync.remote <other>`. So the seam now MARKS the member as
|
|
1596
|
+
dispatched-since-verified and drops nothing; the fingerprint is bound to
|
|
1597
|
+
the URL it was minted against; and a marked member's `sync.remote` is
|
|
1598
|
+
re-read (one `bd config get`, a plain config.yaml read, no Dolt engine)
|
|
1599
|
+
only at the one decision a stale memo could turn into stale data -- the
|
|
1600
|
+
moment a D-pull is about to be SKIPPED on that fingerprint. Same URL:
|
|
1601
|
+
skip. Different or unreadable: fingerprint forgotten, real pull. Every
|
|
1602
|
+
other consumer of the memo fails closed anyway (a wrong "configured"
|
|
1603
|
+
answer just issues a real pull/push against whatever remote bd itself has
|
|
1604
|
+
configured). The re-read is therefore paid once per skip-after-dispatch,
|
|
1605
|
+
never per dispatch; a bracket whose tip moved pays nothing extra; a
|
|
1606
|
+
member with a remote is back to one probe per process. The one case that
|
|
1607
|
+
never reaches that check is a member memoized as having NO remote (both
|
|
1608
|
+
pre-gates exit on it first): its memo is re-read once after any dispatch,
|
|
1609
|
+
because an agent wiring a remote mid-dispatch would otherwise leave every
|
|
1610
|
+
later D-push of that member reporting a benign no-remote skip -- success
|
|
1611
|
+
-- while its bead closes never left the clone. That costs one config.yaml
|
|
1612
|
+
read per dispatch for no-remote members only (sandboxes; the no-remote
|
|
1613
|
+
mock sprint's golden transcript shows one probe per dispatch for exactly
|
|
1614
|
+
this reason).
|
|
1615
|
+
- **The probe never waits on a prompt, and switches itself off when it
|
|
1616
|
+
cannot work.** The probe URL carries no userinfo (9.3), so a member whose
|
|
1617
|
+
only credential for the host was URL-embedded would have git try to
|
|
1618
|
+
PROMPT -- under Git Credential Manager on Windows, a dialog that blocks
|
|
1619
|
+
until the 30s probe timeout, on every bracket. The probe runs with
|
|
1620
|
+
`-c credential.interactive=never -c core.askPass=` (config flags, never a
|
|
1621
|
+
shell env prefix -- the member's shell may be PowerShell) so it fails at
|
|
1622
|
+
once, and after two consecutive failed probes the member's probe is
|
|
1623
|
+
disabled for the process (re-armed by the hard seams), so every pull is
|
|
1624
|
+
simply real again. Two rather than one so a single blip does not cost the
|
|
1625
|
+
feature.
|
|
1626
|
+
|
|
1627
|
+
### 9.2 The transient retry ladder is time-boxed, not count-boxed
|
|
1628
|
+
|
|
1629
|
+
The ladder was widened to 8 retries with a 30s backoff cap to survive a live
|
|
1630
|
+
Windows machine-wide `git.exe` spawn outage (`fork/exec ... "Not enough memory
|
|
1631
|
+
resources"`), measured at 1-3 minutes. That was the right diagnosis but the
|
|
1632
|
+
wrong bound: it applied the long budget to EVERY transient kind, and the unit
|
|
1633
|
+
suite went from 80s to 6m15s because ordinary transients in fixtures now
|
|
1634
|
+
walked the full ladder (91.5s of pure sleep in one exhausted ladder, before
|
|
1635
|
+
counting the 600s per-attempt timeout).
|
|
1636
|
+
|
|
1637
|
+
The budget was never really about a count -- it was about spanning a
|
|
1638
|
+
wall-clock outage window. So the ladder is split by error class:
|
|
1639
|
+
|
|
1640
|
+
| class | bound | backoff cap |
|
|
1641
|
+
|---|---|---|
|
|
1642
|
+
| spawn outage (`fork/exec`, "Not enough memory resources") | 3 min WALL CLOCK | 30s |
|
|
1643
|
+
| every other transient | 5 retries (`maxTransientRetries`) | 8s |
|
|
1644
|
+
|
|
1645
|
+
A wall-clock bound is the same 3 minutes whether each attempt returns
|
|
1646
|
+
instantly or sits on the 600s step timeout; the old count-based bound
|
|
1647
|
+
multiplied out to a worst case of 9 attempts x 600s. The generic default is
|
|
1648
|
+
the pre-widening 5 (a first cut mis-restored it as 2; corrected). An
|
|
1649
|
+
explicitly passed `maxTransientRetries` is honored by BOTH ladders: it bounds
|
|
1650
|
+
the generic class as before, and it is a hard attempt cap on the spawn-outage
|
|
1651
|
+
ladder too (the wall-clock budget still applies underneath it). Left unset --
|
|
1652
|
+
the production shape, since no runner call site passes one -- the spawn-outage
|
|
1653
|
+
class keeps its full wall-clock budget. A `diverged` classification is still
|
|
1654
|
+
never retried, under either ladder. The spawn-outage sub-classifier matches
|
|
1655
|
+
only the `fork/exec` line: the "Not enough memory resources" wording on its
|
|
1656
|
+
own never classifies `transient` in the first place (it is not in the dolt
|
|
1657
|
+
provider's TRANSIENT table), so a pattern for it was dead and was removed.
|
|
1658
|
+
|
|
1659
|
+
### 9.3 Remote-tip fingerprint: skip a D-pull only when it is provably a no-op
|
|
1660
|
+
|
|
1661
|
+
**The correctness anchor.** The shared remote's `refs/dolt/data` is the ONLY
|
|
1662
|
+
channel through which beads state moves between machines. A member's clone can
|
|
1663
|
+
therefore be stale in exactly one way: that ref advanced since the member last
|
|
1664
|
+
pulled or pushed. "Is a D-pull needed?" then has a cheap, EXACT answer --
|
|
1665
|
+
compare the remote SHA now against the SHA this member last synchronized to --
|
|
1666
|
+
rather than a policy bet about who else might be writing.
|
|
1667
|
+
|
|
1668
|
+
This is why the alternatives were rejected. "Skip for read-only roles" and
|
|
1669
|
+
"skip within N seconds of the last pull" both trade real staleness for speed:
|
|
1670
|
+
a reviewer reads acceptance criteria a remote doer just pushed, and a
|
|
1671
|
+
plan-reviewer reads the planner's freshly pushed DAG. Any fixed bound is a
|
|
1672
|
+
guess about other writers, and a wrong guess produces exactly the
|
|
1673
|
+
false-FAILED-streak failure the post-streak verify pull exists to prevent. A
|
|
1674
|
+
tip-equal skip removes only pulls that are provably no-ops, so no operator has
|
|
1675
|
+
to pick a risk bound at all.
|
|
1676
|
+
|
|
1677
|
+
**Mechanism.** Before a real `bd dolt pull`, one `git ls-remote <sync.remote>
|
|
1678
|
+
refs/dolt/data` -- one network round trip, no Dolt engine startup, no 548 MB
|
|
1679
|
+
chunk store to open (0.35s measured, against multi-minute real pulls). On a
|
|
1680
|
+
match the pull is not spawned and the step returns `{ ok: true, member,
|
|
1681
|
+
skipped: true, reason: 'remote-unchanged', remoteTip }`.
|
|
1682
|
+
|
|
1683
|
+
**The recorded tip is minted in exactly one place, from a value the clone is
|
|
1684
|
+
provably current with:**
|
|
1685
|
+
|
|
1686
|
+
- after a successful PULL, record the SHA observed IMMEDIATELY BEFORE it --
|
|
1687
|
+
never one read afterwards. A push racing in during the pull is not in what
|
|
1688
|
+
was fetched, so recording the later SHA would claim freshness the clone does
|
|
1689
|
+
not have. Recording the earlier one can only cost one redundant future pull.
|
|
1690
|
+
- after a successful PUSH by this member, FORGET the recorded tip. A push never
|
|
1691
|
+
records anything.
|
|
1692
|
+
|
|
1693
|
+
**Why a push cannot mint a fingerprint.** Two earlier cuts recorded a SHA read
|
|
1694
|
+
from the remote after the push -- unconditionally at first, then only when the
|
|
1695
|
+
ref had "advanced" past a pre-push read. Both lose the same race: the push
|
|
1696
|
+
mutex serializes this fleet's pushes against each other and nothing else, so
|
|
1697
|
+
an unrelated machine can push in the gap between our push landing and our
|
|
1698
|
+
post-push `ls-remote` returning. That read then observes a stranger's commit
|
|
1699
|
+
this clone has never seen, records it as ours, and the next D-pull is wrongly
|
|
1700
|
+
skipped -- exactly the staleness the feature exists to prevent. A network read
|
|
1701
|
+
of "the remote's current tip" is never a proxy for "what my push published".
|
|
1702
|
+
Nor is there a local proxy for this transport: bd's git-backed remote is Dolt's
|
|
1703
|
+
git blobstore, where a push wraps the chunk-store manifest in a fresh git
|
|
1704
|
+
commit built inside the push on top of the remote head it just fetched. That
|
|
1705
|
+
commit -- the remote's new `refs/dolt/data` -- is not a function of local Dolt
|
|
1706
|
+
state, is not printed by `bd dolt push`, and lands locally only in a
|
|
1707
|
+
per-process UUID-named ref (`refs/dolt/blobstore/<remote>/dolt/data/<uuid>`)
|
|
1708
|
+
inside a sha256-named cache dir under `.beads/`, reachable only by a
|
|
1709
|
+
shell-dialect-specific scan of the member's disk (verified against a live
|
|
1710
|
+
clone: the cache repo's tracking refs and `FETCH_HEAD` already disagreed with
|
|
1711
|
+
the remote's live tip). So the pusher forgets its tip and pays one real pull
|
|
1712
|
+
at its next bracket, after which the skip is re-armed. A no-op push is treated
|
|
1713
|
+
identically -- it proves nothing about the clone's freshness. As a side
|
|
1714
|
+
effect no `ls-remote` is issued anywhere in the D-push bracket: the mutex hold
|
|
1715
|
+
is exactly push + reconcile + settle.
|
|
1716
|
+
|
|
1717
|
+
**Fail-open, always.** Every uncertainty falls through to a REAL pull: no
|
|
1718
|
+
recorded tip yet, an `ls-remote` failure or timeout, unparseable output, a
|
|
1719
|
+
`sync.remote` that could not be positively read, or a URL that fails the
|
|
1720
|
+
strict safe-charset gate. A divergence, a settle, a successful push, an auth
|
|
1721
|
+
self-heal, a `repair()`, an orchestrator-issued `bd init`/`bootstrap`/`config
|
|
1722
|
+
set`, and a `sync.remote` found changed under a dispatch (9.1) all FORGET the
|
|
1723
|
+
recorded tip rather than leave a stale one. The recorded tip is bound to the
|
|
1724
|
+
URL it was observed at and never validates a probe of any other URL. There
|
|
1725
|
+
is deliberately no path in which doubt produces a skip.
|
|
1726
|
+
|
|
1727
|
+
**Three implementation constraints worth restating:**
|
|
1728
|
+
|
|
1729
|
+
- The probe target is resolved from `sync.remote` (through the 9.1 memo, so it
|
|
1730
|
+
costs no extra `bd config get`), NEVER from git's `origin`. The two can
|
|
1731
|
+
legitimately differ on a member, and probing `origin` would compare this
|
|
1732
|
+
member's freshness against the wrong ref. bd spells a git-transport remote
|
|
1733
|
+
`git+https://...`; the `git+` prefix is bd's own scheme marker and is
|
|
1734
|
+
stripped before `ls-remote`. A `sync.remote` with NO scheme -- a bare Dolt
|
|
1735
|
+
remote NAME (`origin`, which the integration fixtures legitimately set) or
|
|
1736
|
+
a bare path (a Dolt file remote, never a git repository) -- yields no probe
|
|
1737
|
+
at all, because `git ls-remote origin` would silently resolve against
|
|
1738
|
+
git's origin, the exact target this rule forbids.
|
|
1739
|
+
- The probe command string is journaled VERBATIM into the persisted,
|
|
1740
|
+
dashboard-visible run transcript (`silent` only suppresses the console
|
|
1741
|
+
line). An http(s) userinfo (`user:token@`) in `sync.remote` is therefore
|
|
1742
|
+
stripped before the URL is interpolated into the command; previously such
|
|
1743
|
+
a token only ever appeared as command OUTPUT (`bd config get`). The
|
|
1744
|
+
stripped URL still authenticates through the git credential helper every
|
|
1745
|
+
provisioned member carries (that is how `git push` works on members); a
|
|
1746
|
+
member whose only credential was URL-embedded gets a failed probe, which is
|
|
1747
|
+
a real pull. ssh URLs keep their `git@` -- a login name, not a secret.
|
|
1748
|
+
- The probe string reaches the member's own shell, which may be PowerShell or
|
|
1749
|
+
a POSIX shell. Rather than trying to quote correctly for both, any remote URL
|
|
1750
|
+
containing a character outside a conservative safe charset (no whitespace,
|
|
1751
|
+
no quote, no metacharacter of either dialect) is REFUSED -- which yields no
|
|
1752
|
+
fingerprint and a real pull, the safe direction.
|
|
1753
|
+
|
|
1754
|
+
Disable per call site with `remoteTipFingerprint: false`.
|
|
1755
|
+
|
|
1756
|
+
### 9.4 Not done here, on purpose
|
|
1757
|
+
|
|
1758
|
+
Two larger items from the same review are deliberately out of scope: squashing
|
|
1759
|
+
the Dolt history and re-bootstrapping every member clone (the largest per-op
|
|
1760
|
+
win, but a destructive operational change needing a quiescent window and its
|
|
1761
|
+
own runbook), and moving `sync.remote` off the git transport onto a native
|
|
1762
|
+
Dolt bucket remote.
|
|
1763
|
+
|
|
@@ -703,7 +703,7 @@ ENGINE's maintainers without silent uploads -- and its flow is adopted whole:
|
|
|
703
703
|
|
|
704
704
|
- **Sanitize FIRST, locally, always** -- before the report is even written to disk: absolute
|
|
705
705
|
paths -> stable placeholders (`<work-folder>/...`, `<home>/...`); hostnames/usernames/IPs/
|
|
706
|
-
emails -> redacted tokens; anything matching secret patterns (keys, tokens, `{{
|
|
706
|
+
emails -> redacted tokens; anything matching secret patterns (keys, tokens, `{{secret.*}}`
|
|
707
707
|
values, env dumps) -> stripped outright, never placeholdered; target-repo identifiers (repo
|
|
708
708
|
name, branch names, bead titles) -> anonymized to `<target-repo>`/`<sprint-branch>` by
|
|
709
709
|
default, with an opt-in tier for internal users to keep them; code snippets excluded by
|