anthropics/financial-services

This system is a plugin-based agent ecosystem designed to automate specialized financial and administrative workflows through modular, version-controlled components. It provides a suite of managed agents for tasks such as earnings analysis, reconciliation, KYC screening, and market research, alongside an administrative tool for configuring cloud-based Office add-in integrations. The infrastructure supports this ecosystem with automated deployment, strict validation, and orchestration scripts that manage cross-agent handoffs and dependency resolution.

This change introduces the \claude-for-msft-65-install\ plugin, an administrative toolset for configuring the Claude Office add-in to connect to custom cloud backends (Vertex AI, Bedrock, Azure AI Foundry, or LLM gateways) instead of Anthropic's default API. The plugin provides CLI commands to generate customized add-in manifests, handle Azure admin consent for Entra SSO and Outlook Graph access, and manage per-user configuration via Microsoft Graph extension attributes. It also introduces a bootstrap endpoint mechanism for delivering dynamic, per-user structured configuration (such as MCP servers and skills) and includes read-only scripts to export user chat history and settings from local browser storage before machine rebuilds or cache clears, ensuring data is preserved since no server-side copy exists.

claude-for-msft-365-install · high confidence

Added five new agent plugins—earnings-reviewer, gl-reconciler, kyc-screener, market-researcher, and meeting-prep-agent—each with a plugin manifest, agent definition, and specialized skills. These plugins enable automated earnings analysis and model updates, GL-to-subledger reconciliation with root-cause tracing, KYC/AML screening and risk rating, sector/thematic market research with competitive analysis, and client meeting briefing packs. Common capabilities include headless Excel (.xlsx) and PowerPoint (.pptx) file generation, integration with internal and third-party MCPs (FactSet, CapIQ, Daloopa, screening, CRM), and strict guardrails treating external documents as untrusted data.

plugins/vertical-plugins · high confidence

This change introduces a suite of new scripts to manage and validate the repository's plugin and managed-agent ecosystem. The \check.py\ script validates YAML/JSON syntax, ensures agent documentation has required frontmatter, verifies that file references in agent manifests resolve correctly, and detects drift between bundled agent skills and their vertical-plugin sources. The \deploy-managed-agent.sh\ script automates the deployment of managed agents by resolving manifest conveniences (inlining system files, uploading skills, creating subagents) and posting the final payload to the Anthropic API. The \orchestrate.py\ script provides a reference event loop for cross-agent handoffs, including security hardening via target-agent allowlisting and payload schema validation. Additional utility scripts include \sync-agent-skills.py\ for re-syncing bundled skills from source, \test-cookbooks.sh\ for dry-running deployments to assert well-formed output, \validate.py\ for harness-side schema validation of worker output, and \version\_bump.py\ to enforce semantic version bumps on modified plugins via pre-commit hooks and CI checks.

scripts · high confidence

A new pre-commit hook has been added to automatically patch-bump any plugin that has staged changes, ensuring the plugin version is exactly one patch level ahead of the main branch. This automation runs the \scripts/version\_bump.py\ script during the commit process, provided Python 3 is available, and can be bypassed for individual commits using the \--no-verify\ flag.

.githooks · high confidence

The \extract\_numbers.py\ script, previously located in the financial-analysis plugin, has been moved to the pitch-agent plugin under \plugins/agent-plugins/pitch-agent/skills/ib-check-deck/scripts/\. This change also corrects a logic error in the \normalize\_number\ function: it now iterates over unit keys sorted by length in descending order to ensure longer unit strings (like 'billion') are matched before shorter ones (like 'bn'), preventing incorrect normalization of financial figures.

python · high confidence

New requirements.txt files have been introduced to define Python dependencies for specific components. The python-bootstrap example now requires fastapi, uvicorn, and PyJWT\[crypto\]. The DCF Model Builder skill, available in both the pitch-agent and model-builder plugin paths, specifies openpyxl (\>=3.0.0) for Excel handling and requests (\>=2.28.0) for HTTP operations. Notably, the model-builder version of the DCF model has removed the yfinance dependency in favor of the generic requests library.

(dependencies) · high confidence

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

• First survey — no prior run to compare against. CAI 57.

• Code Health 83
• Architecture 99
• Maturity 55
• Readiness 41
• Security 94

• 71 commits — 63 feature/other, 8 fixes

• claude-in-office/commands — 15 commits
• (repo) — 13 commits
• claude-for-msft-365-install/commands — 8 commits
• financial-analysis/skills — 7 commits
• (root) — 4 commits
• plugins/vertical-plugins — 4 commits
• claude-for-msft-365-install/.claude-plugin — 3 commits
• .claude-plugin/marketplace.json — 2 commits
• claude-for-msft-365-install/scripts — 2 commits
• claude-in-office/scripts — 2 commits
• spglobal/skills — 2 commits
• .github/workflows — 1 commit
• claude-for-financial-advisors/skills — 1 commit
• claude-in-office/examples — 1 commit
• equity-research/skills — 1 commit
• lseg/commands — 1 commit
• partner-built/lseg — 1 commit
• private-equity/commands — 1 commit
• private-equity/skills — 1 commit
• scripts/deploy-managed-agent.sh — 1 commit

• fix: Fix normalize_number matching 'thousand' as trillion
• fix: Merge pull request #29 from anthropics/fix/normalize-thousand-trillion-substring
• fix: fix(ci): plugin-validate cache fix + msft-365 README update (#239)
• fix: fix(manifest): accept gateway_auth_header and gateway_api_format case-insensitively (#65)
• fix: fix(manifest): add otlp_endpoint/otlp_headers to KEYS validator (#63)
• fix: fix(plugins): add SKILL.md frontmatter, repair hooks.json, dedupe spglobal MCP (#243)
• fix: fix(plugins): repair broken hooks.json + add plugin-validate CI (#232)
• fix: fix: Update docs about Foundry (#323)
• change: Add LSEG Financial Analytics plugin
• change: Add S&P Global plugin
• change: Add SRI integrity hashes to CDN script tags in S&P Global report template
• change: Add ai-readiness skill to private-equity plugin
• change: Add claude-in-office plugin for Office add-in deployment
• change: Add clean-data-xls and audit-xls skills
• change: Add displayName to all marketplace plugin entries (#247)
• change: Add formulas-over-hardcodes guidance and consolidate audit skills
• change: Add read-only add-in data export; harden cache-clear scripts (#302)
• change: Add sovereign / national cloud (GCC-High, DoD, 21Vianet) docs and graph_cloud validation (#268)
• change: Adding Box to MCP Integrations (#187)
• change: Drop claude-for-financial-advisors marketplace entry (#351)
• …and 51 more

• 0 containers · 1 bounded contexts · 0 dependency edges (baseline)

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

anthropics/financial-services was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.