Glossary

adversarial synergy pass

Design review that stress-tests a batch of feature specs in two dimensions: cross-feature consistency and grounding against the real codebase. See How Apogee Works.

backport check

Assessment of whether a function can be safely moved from one branch to another. Evaluates body identity, caller/callee compatibility, and prerequisite presence. See CPG Engine.

body_hash

Blake3 content fingerprint of a function body. Detects meaningful changes vs. whitespace-only edits during incremental CPG builds. See apogee-tree.

branch divergence

Structural comparison between two branches showing functions unique to each side, diverged implementations, and safety flag differences. See CPG Engine.

completion contract

Verification that a chain step actually did its work before advancing. Checks that the required skill was invoked and expected artifacts exist. See Perigee.

constraint routing

Intent classification on every user message. Read mode limits the agent to discovery and inspection. Write mode unlocks the full skill catalog. See User Journey.

delta

Lightweight change record capturing structural differences between two CPG snapshots without duplicating unchanged data. See CPG Engine.

feature slug

Kebab-case identifier for a feature under development (e.g., auth-middleware). Used as the directory name under apogee_artifacts/features/.

full snapshot

Complete CPG build of a branch HEAD, capturing all nodes and edges at that point in time. Contrasted with delta and skeleton builds.

governance gate

Policy-driven checkpoint at a lifecycle boundary. Gates evaluate conditions and return a severity verdict: blocker, warning, acceptable, or informational. See Gate System.

immutable gate

A governance gate that cannot be overridden by policy exceptions. Used for compliance-critical boundaries. See Gate System.

incremental branch loading

SQL-copy strategy that bulk-copies unchanged CPG data from a previously indexed branch, then applies only the delta. See apogee-manifest.

lifecycle chain

Ordered sequence of workflow steps with checkpoints, conditions, failure policies, and resume capability. Chain types: plan, feature, series, tier, release, finalize, workflow. See Chain Engine.

MCP

Model Context Protocol — open standard for connecting AI assistants to external tools. Apogee exposes skills, CPG analysis, and governance as an MCP server over stdio transport. See apogee-mcp.

overlay

Working-directory state layered on top of a committed CPG snapshot. Captures uncommitted changes for real-time analysis without modifying the persisted graph.

per-subsystem history

Path-scoped CPG indexing that tracks structural evolution within a declared subsystem boundary, enabling targeted temporal queries.

Perigee

OTP-supervised chain execution engine. Runs lifecycle chains as isolated, crash-recoverable GenServer processes with completion contracts and agent transfer. See Perigee.

plan scaffolding

Generating plans.md, plan_prompts/PLAN-0X.md, and review_prompts/ for a feature before implementation begins.

proof depth

Rigor level of a verification pass: smoke (basic health), functional (feature-level), adversarial (stress/edge-case), destructive (fault injection). See apogee-proof.

proof harness

Release verification framework that runs structured test suites and produces machine-readable proof reports with per-surface verdicts. See apogee-proof.

risk score

Numeric score ranking functions by security-relevant traits: unchecked input, unsafe operations, complexity, dead code. See CPG Engine.

scope_summary

Per-function safety profile with 13 boolean flags (has_allocation, has_untrusted_input, etc.). Computed during CPG builds and used by security scanning tools.

security scan

Compound analysis identifying unchecked untrusted input, unsafe copies, complex allocators, dead code, and risk-ranked functions. See CPG Engine.

series

Numbered batch of related features designed and implemented together (e.g., series37-temporal-cpg-model). Each series has an implementation order and tier structure.

skeleton

Reduced CPG build capturing only structural outline (functions, classes, imports) without full body content. Used for fast historical indexing.

skill

Reusable workflow step or task for an AI agent, authored in Markdown and compiled into versioned Python modules. Discovered via find_skill, executed via run_skill. See Skill Runtime.

skill collection

A group of related skills that form a workflow suite (e.g., bug_investigation, code_porting, docs-create). Skills within a collection are linked via next_workflow.

stable identity

Composite key (kind, name, file_path) that uniquely identifies a CPG node across commits and branches. Enables temporal tracking of individual functions and types.

subsystem

Logical code boundary grouping related files for scoped analysis. Auto-detected from MAINTAINERS, CODEOWNERS, workspace files, or top-level directories. Persisted to .apogee/subsystems.toml. Most CPG tools accept a subsystem parameter to filter queries.

tier

Dependency level within a series. Tier 0 features have no dependencies. Tier 1 depends on Tier 0. Features within a tier can be implemented in parallel.

See Also