Track the evolution of TokenCap from a token-budgeted context generator to a complete AI Project Intelligence Platform.
Version Timeline 3.0.0
Opens TokenCap up as a platform. A dependency-free, read-only SDK (@tokencap/sdk) lets local tools load, search, and query intelligence that tokencap make has already generated. Team intelligence keeps shared rules, decisions, approvals, profiles, and ADRs as append-only records in Git under .tokencap/team/. Projects can opt into local plugins with after-build and before-redact hooks confined to their own artifact directory. tokencap make --migrate upgrades a v2 workspace to the v3 artifact contract, and the v2 standalone command aliases are removed: each now prints its canonical replacement.
Makes TokenCap discoverable and executable by any AI agent through the open skill system: a standard skill definition (SKILL.md, commands, workflows) for OpenCode, Claude Code, Cursor, Codex, and Antigravity ships inside the npm tarball and installs through each host's own skill-discovery mechanism without overwriting host configuration. Adds workflow playbooks for onboarding, refactoring, impact audits, and change reviews, and stabilizes MCP Contract v2 with a fixed tool contract, generated artifact schemas, and runtime artifact validation.
Ships the TokenCap Companion workflow for ChatGPT, Claude, and Gemini: a local, paired bridge that injects bounded redacted project context without replacing drafts, supports preview/copy/download recovery, and stages browser captures for explicit approval. It also consolidates the CLI into seven core commands while keeping legacy aliases compatible.
Closes the conversation-to-repository loop with Session Capture v1 and staged inbound memory; introduces 0–100 repository build health diagnostics and explainable scoring, normalized evidence-rich risk findings with PEP 621/Poetry python parsing, persistent watcher controls with session ROI tracking, and a cold-start project primer for browser companions.
Makes the local loopback bridge fast, model-aware, and always-on: caches project intelligence with 10s warm reuse (repeat calls drop ~3x from 2.6s to 0.8s), adds a true cache-clearing /refresh rebuild, dynamically budgets task packs based on target LLM context windows, and introduces background daemon lifecycle management via 'tokencap serve --daemon' and '--stop'.
Introduces 'tokencap serve', a loopback-only HTTP bridge on 127.0.0.1 implementing Bridge Contract v1 (health, projects registry, query-scoped and budget-managed /generate, /refresh, and /capture 501 stub) with pairing token auth and Origin allowlisting — enabling the sandboxed Chrome Companion extension to access local repository intelligence with zero remote network egress.
Promotes Go, Rust, and Java to full Tier-1 call-graph resolution; adds Vue/Svelte/Angular SFC component intelligence; introduces 'tokencap analyze schema' for SQL & GraphQL schema extraction with code links; adds 'tokencap analyze taint' for intra-procedural JS/TS taint flow analysis from sources to sinks; and ships grammars inside dist/ so the package no longer requires src/.
Introduces Guided Safe Refactoring ('tokencap refactor rename' and 'tokencap refactor rm-dead') with byte-exact transform guarantees and refuse-on-doubt protections, Static Test Mapping & Gap Analysis ('tokencap analyze tests'), coverage report ingest, and local-only telemetry-free Agent Feedback tracking ('tokencap feedback').
Introduces 'tokencap analyze review' for evidence-backed local review packets, 'tokencap analyze ownership' for opt-in churn and collaborator context, and selected host pointer scoping via 'tokencap mcp --init --client <host>'.
Introduces 'tokencap analyze simplify' for high-confidence unreachable code and duplicates, tracking TODO/DEBT markers with local first-seen dates via '.tokencap/debt/ledger.json', and making 'tokencap compress' dry-run by default.
Tree-sitter symbol analysis across seven languages, symbol-level impact and dead-code detection, frontend intelligence, durable ADRs, bounded parallel graph builds, memory ceilings, atomic sharded caching, workspace federation, package-scoped MCP queries, infrastructure dependency nodes, and byte-exact markdown compression.
Official host plugins for Claude Code & Codex (tokencap-plugin), restructured subcommand hierarchy (tokencap ask brain, tokencap analyze diff, tokencap security scan), VS Code Extension auto-capture, and auto-inferred memory notes.
Real-time Token & Dollar Savings Engine, 12.2× benchmark reduction (94.3% cost savings), multi-host pointer generation (AGENTS.md, CLAUDE.md, Cursor, Windsurf, Cline, Copilot, Kiro), and CI-enforced zero-network egress guarantees.
Automatic build-on-startup for MCP, live repository background watcher for source & notes changes, universal host setup (tokencap mcp --init --client <host>), safe config merging, and zero-latency in-memory caching.
Local Model Context Protocol (MCP) support with 11 stdio tool endpoints, auto IDE configuration (tokencap mcp --init), centralized secret redaction engine, and up to 92% token budget savings.
Analyze once. Update only what changed. SHA-256 hash cache, bidirectional dependency walk, per-cluster selective rebuilds, delta files, and versioned history. Sub-100ms for single-file edits.
Introduces the Repository Constitution (tokencap constitution) to encode permanent engineering laws regarding API contracts, schemas, security, and compliance.
Introduces the Execution Contract (--execution) for engineering discipline: Scope Drift Detection, Minimal Solution First questions, Confidence Rating, and Recovery Mode.
Launches the Self-Loading Universal Context Layer (AGENTS.md, START_HERE.md, allowed-context.json, context-manifest.json), topic-based pack compression, and future IDE exporter configurations.
Introduces Project Brain for unified cluster-based intelligence query with fuzzy matching and 6 mode flags.
Analyzes repository to generate AI Agent Skill Packs (agent-pack.md) and machine-readable definitions.
Question-based context retrieval with Graph BFS, topic filters, and token budget management.
Obsidian-style visual dependency graph, node risk scoring, and unified CLI command suite.
Interactive HTML Cytoscape graph viewer, narrative AI summaries, and token-budgeted context packing.
AI-powered semantic change analysis, breaking change detection, and API endpoint mapping.
CLI-based debugging memory subsystem to capture environment, failed tests, and stack traces.
Dependency graph mapping (TOKENCAP_GRAPH.md) and developer context memory tracker.
Auto-save snapshots, context profiles, secret redaction, and integrated VS Code status bar.