Session Metadata Schema in Council of High Intelligence: Complete Field Reference

The Session Metadata schema tracks eleven telemetry fields—including execution mode, panel composition, token estimates, and fallback triggers—that the coordinator appends to every verdict in a grep-friendly YAML front-matter block.

The Council of High Intelligence coordinator appends a standardized Session Metadata block to every verdict generated by the council. This schema version 1 implementation, defined in SKILL.md, provides a fixed telemetry contract that downstream tooling, log aggregators, and validation pipelines can reliably parse from markdown output.

Core Telemetry Fields

The Session Metadata schema captures operational data across three categories: session configuration, runtime behavior, and resource consumption. According to the specification in SKILL.md (lines 41-51), the coordinator populates the following fields:

Session Configuration

• schema_version – Identifies the metadata format revision. Currently locked to 1 to ensure parser compatibility.
• mode – Indicates the execution mode employed: full, quick, duo, or triad. This determines the deliberation depth and consensus requirements.
• panel_size – The integer count of council members that participated in the deliberation round.
• rounds_run – The actual number of deliberation iterations executed before reaching a verdict.

Operational Behavior

• chairman_failed_fallback – Boolean indicator (yes/no) showing whether the Chairman fallback logic triggered due to primary model failures.
• tools_used – Flag (yes/no) indicating whether auxiliary tools such as file read, grep, or URL fetch were invoked during the session.
• fallbacks_triggered – Comma-separated list of provider or model substitutions required during execution, or the string "none" if all primary models responded successfully.

Resource Utilization

• input_tokens_estimate – Approximate token count processed as input (e.g., ~12k), useful for cost estimation and capacity planning.
• output_tokens_estimate – Approximate token count generated in the response.
• duration_seconds – Rough wall-clock runtime of the session (e.g., ~30).
• provider_count – Integer derived from detect-providers.sh representing the number of distinct model providers participating in the council.

Schema Format and Validation

The Session Metadata block follows a strict markdown-native format. As documented in SKILL.codex.md (lines 42-44), the block uses YAML front-matter delimited by --- to enable easy redirection to log files and straightforward extraction via standard Unix tools.

The validation script scripts/council-simulation-checklist.sh enforces this contract. Lines 106-109 assert the presence of the metadata block and validate that schema_version: 1 appears exactly as specified, failing the CI check if the coordinator omits or malforms the telemetry.

Extracting Session Metadata

Because the schema uses standard YAML front-matter syntax, you can extract metadata from any verdict file without specialized parsers.

To isolate the block from a verdict:

grep -A12 '^schema_version:' verdict.md

To append metadata to a structured log:

grep -A12 '^schema_version:' verdict.md >> session-telemetry.log

Example Metadata Block

When the coordinator emits a verdict, it appends a block matching this structure:

---
schema_version: 1
mode: quick
panel_size: 3
rounds_run: 2
chairman_failed_fallback: no
tools_used: yes
input_tokens_estimate: ~8k
output_tokens_estimate: ~5k
duration_seconds: ~12
provider_count: 2
fallbacks_triggered: none
---

In a full verdict context, this appears after the deliberation content:

## Full Council Verdict

### Problem

Should we add caching to the authentication layer?

### Vote Tally

...
---

schema_version: 1
mode: full
panel_size: 5
rounds_run: 3
chairman_failed_fallback: yes
tools_used: yes
input_tokens_estimate: ~15k
output_tokens_estimate: ~10k
duration_seconds: ~30
provider_count: 3
fallbacks_triggered: member→anthropic/claude-2, member→openai/gpt-4

Summary

• The Session Metadata schema defines eleven standardized fields tracking configuration, operations, and resource usage.
• Schema version 1 is enforced by scripts/council-simulation-checklist.sh (lines 106-109) to guarantee parser compatibility.
• The format uses YAML front-matter delimiters for grep-ability and log aggregation.
• Key source files include SKILL.md (primary specification), SKILL.codex.md (Codex mode variant), and the validation shell script.

Frequently Asked Questions

How do I extract Session Metadata from generated verdicts?

Use standard grep with context lines to capture the YAML block. The command grep -A12 '^schema_version:' verdict.md reliably extracts the entire metadata section because the schema mandates the schema_version field appear first in the block.

What happens if the schema_version field is missing?

The validation script scripts/council-simulation-checklist.sh will fail the session check. Lines 106-109 explicitly assert the presence of schema_version: 1, ensuring that any coordinator output missing this marker gets rejected in CI pipelines.

Which values are valid for the mode field?

The specification recognizes four execution modes: full (complete deliberation), quick (abbreviated consensus), duo (two-member panel), and triad (three-member panel). These values are documented in SKILL.md line 42 and affect how panel_size and rounds_run are interpreted.

How is the provider_count field calculated?

The coordinator derives this value from the output of detect-providers.sh, counting the number of distinct model providers (e.g., OpenAI, Anthropic, local inference) that successfully participated in the session, as specified in SKILL.md line 50.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}