bicameral-capture-corrections

Detect missed user corrections and ingest them as proposed decisions via bicameral.ingest.

Updated Aug 27, 2026
One-click install
npx skills add https://github.com/BicameralAI/bicameral-mcp --skill bicameral-capture-corrections
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: bicameral-capture-corrections
Source: https://github.com/BicameralAI/bicameral-mcp/tree/main/.claude/skills/bicameral-capture-corrections
Command: npx skills add https://github.com/BicameralAI/bicameral-mcp --skill bicameral-capture-corrections

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Scans for corrections stated mid-session that were not captured in the decision ledger, ensuring load-bearing design, scope, or constraint decisions are recorded. Automatically fires at session end via the SessionEnd hook and can also be invoked manually after a session with implicit decisions.

Core Features & Use Cases

  • In-session detection of corrections and silent ingestion of mechanical fixes
  • SessionEnd batch on exit to surface unrecorded corrections for review
  • Ingest corrections as proposals for later ratification

Quick Start

Run preflight during a session to surface corrections and rely on SessionEnd to ingest any unrecorded decisions.

Frequently Asked Questions about bicameral-capture-corrections

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I capture mid-session corrections that were missed in the decision ledger?

To capture mid-session corrections, the Skill scans recent turns for unrecorded decisions and silently ingests mechanical fixes, then deduplicates them against the existing ledger using a canonical rubric.

Can I batch ingest unrecorded decisions at the end of a session?

Yes, you can batch ingest unrecorded decisions at session end. The Skill automatically fires via the SessionEnd hook to scan the full transcript and surface corrections for review.

How do corrections get filed into the ledger as proposed decisions?

Corrections are filed as proposed decisions through bicameral.ingest, using a session-based source_ref and naming context to ensure proper classification and traceability for later ratification.

Do I need to manually trigger correction detection during an active session?

No manual trigger is required for end-of-session capture, but you can invoke preflight manually during a session to surface implicit corrections before the SessionEnd hook runs.

How does the deduplication process work when ingesting session corrections?

Deduplication works by applying a canonical rubric to each correction and checking it against the existing decision ledger to prevent duplicate entries before filing them as proposed decisions.

What happens to load-bearing design and scope decisions stated mid-session?

Load-bearing design, scope, and constraint decisions stated mid-session are identified by scanning recent turns and ingested into the ledger as proposals to ensure they are officially recorded.