Skip to content

Connect product workflows, board delivery and retained recovery - #93

Open
ancongui wants to merge 212 commits into
mainfrom
integration/crm-reconciliation-current-main-20261031
Open

ancongui wants to merge 212 commits into
mainfrom
integration/crm-reconciliation-current-main-20261031

Conversation

@ancongui

@ancongui ancongui commented Oct 31, 2026 •

Copy link
Copy Markdown
Contributor

Product workflows keep long forms in the workspace, connect reviewed source inspection to product conversations, and retain uncertain assistant requests across navigation. Project access now uses a compact person search with identifiable results and an explicit selection. Protected consent dialogs adapt to the available space while preserving typed input, keyboard access and visible confirmation actions.

The GitHub Projects workflow adds project-scoped account verification, private board discovery and creation, status mapping, and delivery receipts. Retained-result recovery checks the original claimed action and native owner before local postprocessing; it preserves spent usage and grants no additional coding or tool execution.

Status replies now distinguish current typed checks and the current error from historical worker quotations. The writer receives the classifier draft as untrusted context, preserves the same claimed recovery and verified delivery links, and shares one fact packet across drafting, review and one correction. It adds no execution grant or new draft-length cap.

Documentation follows the account, scope and trust-confirmation steps and includes current screenshots from controlled fixtures.

Validation:

  • Studio: 926 unit tests, TypeScript and production build pass; lint and formatting pass. Its 488 capture inputs remain unchanged after the backend-only status fix.
  • Status communication: 290 host cases on the initial fix and 99 affected cases on the follow-up pass, with real models, SQL, network and forbidden processes disabled. Independent review confirmed the final source and all 366 historical template rows/bytes are preserved.
  • Browser walkthrough: one fresh complete run, 138 passed. All 694 captures are bound to the current frontend, helpers, producers and configuration. The earlier 136/2 result is preserved; its visible-scrollbar defect is fixed and covered.
  • Documentation: both strict MkDocs builds, 22 Python tests, 10 Node tests, six responsive gallery cases and five Pages cases pass. The path check verifies 76 outputs and 11,067 local URLs.
  • The composed backend retains 202 MCP tools and 373 versioned templates. Source-level compatibility checks are separate from the required native runtime qualification.

PR94 was merged normally into PR93. Current head 997b91bd0909784762dfa629fa21393b8b46292e has parents 5ed901b and 486696c; its Git tree 49f21c4f50f161afcb7abf182f300d400003949b is byte-exact to qualified source 486696ccb982b8e283813c566d3de7dc9e2edfd2. Fresh CI for 997b91b is pending; no new-head success is claimed. Exact 486696 previously passed full Verify 41333778963 and Documentation 41333779009, actual Native250/251 requests, SQL201 in 12 drained packets, REG34/loader35 registration and installed App22 profile/RPC/native-UoW verification. These runtime proofs remain bound to 486696; tree equality is source-content evidence, not a retargeted qualification or new installed adoption. Native tool-boundary proof does not establish whole-controller production confinement. Main is not merged. Screenshots use synthetic API fixtures and do not establish generated CRM acceptance. Runtime deployments remain local.

Known product boundary: Original same-claim reconciliation is accepted at journal 361. One separate seven-check operation has four known successful commands and is held in needs_reconciliation, unknown=false, with no ordinal 5 intent. The original exception was not recorded; a controlled reproduction identified a selected untracked Vitest cache-file guard. Its source fix and the bot current-facts correction remain isolated and uninstalled. The latest real bot reply contradicted the accepted recovery and known results. No rerun, reset, replacement operation, new coding grant or new CRM commit/PR is claimed; acceptance remains 41/87. Source integration does not assert completed CRM end-to-end acceptance or authorize replay.

Andres Contreras added 30 commits October 30, 2026 01:35
Andres Contreras added 11 commits October 31, 2026 21:25
Require an exact published review and the same verified owner before one fresh linked-verification request. Keep current native host-admin proof separate from the mapped SCM actor; use the same request for uncertain readback and explain current linked results through typed facts.

Preservation cut, not ready for main: the combined host cohort recorded 157 passes and one setup error at the real source-catalog free-space floor. An earlier isolated native-manager consumer passed, without transferring that proof. The final three-file ambiguity guard has Ruff/AST validation but has not been rerun through tests or mypy. No frontend inputs changed.
Andres Contreras and others added 17 commits October 31, 2026 22:22
Revalidate the current pending conversation under native locks and bind its immutable intake digest to the grant. Recover typed current or unknown operation facts when interruption precedes the final reply marker, without repeating admission or substituting parent check history.

Focused host regressions: 48 passed, prohibited effects zero; mypy eight modules and Ruff/format seven files passed. The known native-manager disk-floor fixture remains explicitly unexecuted on this cut. No frontend, API wiring, native service, qualification or runtime adoption change is claimed.
# Conflicts:
#	src/firefly_forge/linked_verifications.py
Append workspace_tools without shifting historical admission materials or changing membership. Keep historical/core hashes intact and declare the exact current additive authority tail. Exercise valid command indices above the old limit while retaining strict SDK refusals before HTTP.

Reproduced all five owned Python CI failures, then passed the same nodes and 46 focused host cases with prohibited effects zero. Twenty-eight unrelated custody cases were not selected. Mypy one module and Ruff/format three files passed. Other hosted failures and fresh composed qualification remain separate.
Report only a closed-whitelist exception kind when initial best-effort observation fails. Preserve production one-second sampling, cancellation and recorded task state. Give the SQL custody fixture an explicit observation budget while keeping every predicate and revocation assertion intact.

Deterministic host regressions reproduce missing diagnostics then pass all 31 observer cases with all six effect counters zero. Mypy one source module and Ruff/format three files pass. Actual PostgreSQL remains unexecuted; the original CI exception kind and timeout hypothesis are not proven.
…ed-current-20261031

Verify retained coding results through an independent native ledger

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant