CLI Reference

Cognition Engines provides two CLI tools:

1. bin/cognition — Local/offline tool for direct database access (server-side)
2. scripts/cstp.py — Remote client for connecting to the CSTP server (client-side)

---

1. Local CLI (bin/cognition)

Location: bin/cognitionUse case: Server maintenance, offline indexing, direct guardrail checks.

Prerequisites

• Python 3.11+
• cognition-engines package installed
• GEMINI_API_KEY environment variable
• ChromaDB accessible

Commands

cognition index <directory>

Index all YAML decision files from a directory into ChromaDB.

# Index decisions from the default directory
python bin/cognition index decisions/

cognition query <context>

Search for semantically similar decisions.

# Basic query
python bin/cognition query "database selection"

Options:

Flag	Description
--category CAT	Filter by decision category
--min-confidence N	Minimum confidence threshold

cognition check

Evaluate guardrails against a decision context.

python bin/cognition check --category architecture --stakes high --confidence 0.8

cognition guardrails

List all loaded guardrail definitions.

cognition count

Count indexed decisions.

cognition patterns <subcommand>

Run pattern analysis: calibration, categories, antipatterns, full.

---

2. CSTP Client (scripts/cstp.py)

Location: scripts/cstp.pyUse case: Agent interaction, recording decisions, querying the server.

Setup

Requires .secrets/cstp.env with CSTP_URL and CSTP_TOKEN.

cstp.py query

Query similar decisions from the server.

python scripts/cstp.py query "how to persist state" --top 5 --mode hybrid

Options:

Flag	Description
--top N	Number of results (default: 5)
--category CAT	Filter by category
--project PROJ	Filter by project
--mode MODE	Retrieval mode: semantic, keyword, hybrid
--bridge-side SIDE	Search by bridge side: structure, function

cstp.py check

Check guardrails before acting.

python scripts/cstp.py check -d "deploy to prod" -s high -f 0.9

cstp.py record

Record a decision. Record early to capture deliberation inputs.

python scripts/cstp.py record \
  -d "Plan: Use PostgreSQL for persistence" \
  -c architecture \
  -s high \
  -f 0.85 \
  --tag database --tag infrastructure \
  --pattern "Choose ACID-compliant stores for transactional data"

Options:

Flag	Description
-d, --decision	Decision text (required)
-c, --category	Category (required)
-s, --stakes	Stakes: low, medium, high, critical
-f, --confidence	Confidence (0.0-1.0)
--context	Context description
-r, --reason	Add reason (type:text, repeatable)
--tag, -t	Reusable keyword tag (repeatable)
--pattern	Abstract pattern this decision represents
--project	Project (owner/repo)
--pr	PR number
--structure	Bridge: structure/pattern
--function	Bridge: function/purpose
--tolerance	Bridge: features that don't matter
--enforcement	Bridge: features that must be present
--prevention	Bridge: features that must be absent

cstp.py think

Record a chain-of-thought reasoning step. Pre-decision mode (no --id) accumulates in the tracker; post-decision mode (--id) appends to an existing decision's trace.

# Pre-decision: captured automatically when you record
python scripts/cstp.py think "Considering Redis vs Memcached for caching"

# Post-decision: appends to existing decision trace
python scripts/cstp.py think --id <ID> "Redis chosen because it supports persistence"

cstp.py update

Update an existing decision's fields. Use after recording to finalize with actual outcomes.

python scripts/cstp.py update <ID> \
  -d "Used PostgreSQL with connection pooling" \
  --context "Deployed with PgBouncer. 3 replicas." \
  --tag database --tag infrastructure --tag pgbouncer

Options:

Flag	Description
-d, --decision	Updated decision text
-f, --confidence	Updated confidence (0.0-1.0)
--context	Updated context
--tag, -t	Tags (repeatable, replaces existing)
--pattern	Abstract pattern

cstp.py pre

Pre-decision helper: runs query + guardrail check in one call. Use before recording.

python scripts/cstp.py pre "deploy retry logic to production" -s high -f 0.85

cstp.py get

Get full decision details by ID, including reasons, bridge, related decisions, and deliberation.

python scripts/cstp.py get <ID>

cstp.py review

Review a decision outcome.

python scripts/cstp.py review --id <ID> --outcome success

cstp.py calibration

Get calibration stats.

python scripts/cstp.py calibration

cstp.py reason-stats

Get reason-type statistics.

python scripts/cstp.py reason-stats --min-reviewed 5