Skip to content

docs: remove confidential internal content from public docs#5

Open
muzaffar-ni wants to merge 2 commits into
mainfrom
sanitize-internal-stack-terms
Open

docs: remove confidential internal content from public docs#5
muzaffar-ni wants to merge 2 commits into
mainfrom
sanitize-internal-stack-terms

Conversation

@muzaffar-ni

@muzaffar-ni muzaffar-ni commented Jul 9, 2026

Copy link
Copy Markdown

Why

This repo is public, and the Transilience AI chat assistant grounds its answers in it. A user probing the assistant ("why are you calling it a claude code agent?") got it to quote these docs and confirm internal implementation details, defeating the confidentiality guardrails in transilienceai/mcs#304. This PR is a deep audit of the whole repo to remove confidential/internal content — not just vendor names.

What changed

concepts/Applications.md — rewritten (825 → ~330 lines). It was effectively an internal engineering reference. It's now a clean, user-facing app catalog: each app keeps its description, purpose, high-level data sources (e.g. "your AWS account"), what you get back (reports/dashboards/Excel), and framework coverage. Removed entirely:

  • The shared Platform Architecture section and all Mermaid diagrams (architecture, per-app pipelines, request-sequence)
  • Internal agent / skill / hook names
  • Internal endpoints (/project/execute, /ai-bridge/invoke|stream)
  • Internal file names (task_results.json, *.tsx components, flow.json)
  • Internal scripts & build tooling (run_pipeline.py, collect_aws_data.py, esbuild, DuckDB, boto3, OIDC/NDJSON, the projects/ + shared/ monorepo)
  • Vendor/model names (Claude, Anthropic, Sonnet, Modal, CLDPM) — the original goal of this PR
  • The empty internal placeholder app

Concept docs & glossary — minor internal-naming asides removed: "project is the internal name for an app", "conversation is the internal name for a thread", "artifact type", "the backend", and the app-detail "agents and skills used internally" bullet.

Audited and left unchanged: README.md, install/*, workflows/*, and the rest of concepts/* — these are pure user-facing UI docs with no internal content.

⚠️ Follow-up needed

Applications.md is auto-generated from projects/*/CLAUDE.md + project.json in claude-compliance-suite. The generator will reintroduce all of this on the next regeneration — it needs to be updated to emit user-facing catalog content (or excluded from the public repo). The mcs frontend now also redacts vendor terms from retrieved docs as a code-level backstop, but that only covers vendor names, not the architecture/diagrams — so the generator fix is the durable one.

🤖 Generated with Claude Code

The AI assistant grounds its answers in this public repo and was quoting
this file to confirm internal implementation details when users probed
("why are you calling it a claude code agent?"). Replace vendor and
infrastructure names with neutral platform terminology:

- "CLDPM-managed Claude Code project" -> "platform-managed agent project"
- "Claude Code Agent(s)" -> "Transilience Agent(s)"
- "Claude API (proxy)" / "Claude Sonnet 4.6" -> "AI engine"
- "Modal OIDC" -> "platform OIDC"
- "CLDPM Monorepo" -> "App Monorepo"
- CLAUDE.md references -> "project config" / "committed configuration"

NOTE: this file is generated from projects/*/CLAUDE.md and project.json
in claude-compliance-suite. The generator needs the same substitutions,
or the next regeneration will reintroduce these terms. (The frontend
also now redacts vendor terms from retrieved docs as a backstop.)
Deep audit of the whole repo for confidential/internal content, since
the AI assistant grounds its answers here and the repo is public.

Applications.md was an internal engineering reference (825 lines). Rewrote
it into a user-facing app catalog: kept each app's description, purpose,
high-level data sources, what you get back, and frameworks. Removed all
internal architecture:
- the shared Platform Architecture section and all Mermaid diagrams
  (architecture, per-app pipeline, and request-sequence diagrams)
- internal agent / skill / hook names
- internal endpoints (/project/execute, /ai-bridge/invoke|stream)
- internal file names (task_results.json, *.tsx components, flow.json)
- internal scripts and build tooling (run_pipeline.py, collect_aws_data.py,
  esbuild, DuckDB, boto3, OIDC/NDJSON, the projects/ + shared/ monorepo)
- the empty internal placeholder app

Also removed minor internal-naming asides from the concept docs and
glossary ("project is the internal name for an app", "conversation is
the internal name for a thread", "artifact type", "the backend", and
the app-detail 'agents and skills used internally' bullet).
@muzaffar-ni muzaffar-ni changed the title docs: remove internal stack/vendor terms from Applications.md docs: remove confidential internal content from public docs Jul 9, 2026
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