kirograph 0.13.0 → 0.14.0

package/README.md

@@ -8,11 +8,13 @@ Semantic code knowledge graph for [Kiro](https://kiro.dev): fewer tool calls, in
 
 Inspired by [CodeGraph](https://github.com/colbymchenry/codegraph) by [colbymchenry](https://github.com/colbymchenry) for Claude Code, rebuilt natively for Kiro's MCP and hooks system.
 
+> **Full support is for Kiro only.** Experimental integrations for other MCP-capable tools (Claude Code, Codex) are available but not fully tested. See [Other Tools (Experimental)](#other-tools-experimental) for details.
+
 ## Why KiroGraph?
 
 When you ask Kiro to work on a complex task, it explores your codebase using file reads, grep, and glob searches. Every one of those is a tool call, and tool calls consume context and slow things down.
 
-KiroGraph gives Kiro a semantic knowledge graph that's pre-indexed and always up to date. Instead of scanning files to understand your code, Kiro queries the graph instantly: symbol relationships, call graphs, type hierarchies, impact radius — all in a single MCP tool call.
+KiroGraph gives Kiro a semantic knowledge graph that's pre-indexed and always up to date. Instead of scanning files to understand your code, Kiro queries the graph instantly: symbol relationships, call graphs, type hierarchies, impact radius, all in a single MCP tool call.
 
 The result is fewer tool calls, less context used, and faster responses on complex tasks.
 
@@ -20,12 +22,12 @@ The result is fewer tool calls, less context used, and faster responses on compl
 
 KiroGraph uses [tree-sitter](https://tree-sitter.github.io/tree-sitter/) to parse your source files into an AST and extract:
 
-- **Nodes** — functions, methods, classes, interfaces, types, enums, variables, constants, routes, components, and more (24 node kinds total)
-- **Edges** — calls, imports, exports, extends, implements, contains, references, instantiates, overrides, decorates, type_of, returns
+- **Nodes**: functions, methods, classes, interfaces, types, enums, variables, constants, routes, components, and more (24 node kinds total)
+- **Edges**: calls, imports, exports, extends, implements, contains, references, instantiates, overrides, decorates, type_of, returns
 
 Everything is stored in a local SQLite database (`.kirograph/kirograph.db`). **Nothing leaves your machine.** No API keys. No external services.
 
-The index is kept fresh automatically via Kiro hooks — no background watcher process needed.
+The index is kept fresh automatically via Kiro hooks when using the Kiro integration; no background watcher process needed.
 
 ## How Indexing Works
 
@@ -53,7 +55,7 @@ These embeddings power natural-language search in `kirograph_context` and act as
 | `qdrant` | `.kirograph/qdrant/` | ANN (HNSW), sub-linear | `qdrant-local` (embedded binary) |
 | `typesense` | `.kirograph/typesense/` | ANN (HNSW), sub-linear | `typesense` (auto-downloaded binary) |
 
-Each engine owns its embedding store exclusively — nothing is written to the SQLite `vectors` table when a non-cosine engine is active. If an engine's optional dependency is not installed, KiroGraph silently falls back to `cosine`.
+Each engine owns its embedding store exclusively; nothing is written to the SQLite `vectors` table when a non-cosine engine is active. If an engine's optional dependency is not installed, KiroGraph silently falls back to `cosine`.
 
 Enable and configure via `kirograph install` (interactive arrow-key menu) or directly in `.kirograph/config.json`:
 
@@ -66,7 +68,7 @@ Enable and configure via `kirograph install` (interactive arrow-key menu) or dir
 
 ### Architecture analysis (opt-in)
 
-When `enableArchitecture: true` is set, KiroGraph detects the high-level structure of your project — packages and architectural layers — and computes coupling metrics between them. Results are stored in `arch_*` tables inside `kirograph.db` and exposed via dedicated MCP tools and CLI commands.
+When `enableArchitecture: true` is set, KiroGraph detects the high-level structure of your project (packages and architectural layers) and computes coupling metrics between them. Results are stored in `arch_*` tables inside `kirograph.db` and exposed via dedicated MCP tools and CLI commands.
 
 Enable via `kirograph install` or directly in `.kirograph/config.json`:
 
@@ -109,15 +111,20 @@ kirograph --version
 ### Remove from a project
 
 ```bash
-kirograph uninit [path]     # Remove .kirograph/, hooks, and steering file from the project
-kirograph uninit --force    # Skip confirmation prompt
+kirograph uninit [path]                  # Prompts to remove Kiro integration files and .kirograph/ data separately
+kirograph uninit --force                 # Remove Kiro integration files + .kirograph/ data without confirmation
+kirograph uninit --target all --force    # Remove all integration files (Kiro + Claude + Codex) + .kirograph/ data
 ```
 
-This removes:
-- `.kirograph/` — index database, snapshots, and export directory
-- `.kiro/hooks/kirograph-*.json` — all KiroGraph hooks
-- `.kiro/steering/kirograph.md` — the steering file
-- `.kiro/agents/kirograph.json` — the CLI agent config
+`kirograph uninstall` is an alias for `kirograph uninit`.
+
+Without `--force`, KiroGraph asks separately whether to remove the selected tool integration files and whether to remove the shared `.kirograph/` data. With `--force`, both are removed unconditionally.
+
+This can remove:
+- `.kirograph/`: index database, snapshots, and export directory
+- Kiro target: `.kiro/hooks/kirograph-*.json`, `.kiro/steering/kirograph.md`, `.kiro/agents/kirograph.json`
+- Claude target (experimental): `kirograph` from `.mcp.json`, plus the KiroGraph import from `CLAUDE.md`
+- Codex target (experimental): the generated KiroGraph block from `AGENTS.md`
 
 ### Remove the CLI globally
 
@@ -138,10 +145,10 @@ npm uninstall -g .
 
 ```bash
 # In your project:
-kirograph install    # wire up MCP + hooks + steering + CLI agent in .kiro/
+kirograph install                  # wire up Kiro MCP + hooks + steering + CLI agent
 ```
 
-Restart Kiro IDE, or switch to the `kirograph` agent in Kiro CLI. It will now use KiroGraph tools automatically.
+All Kiro integration files are written to `.kiro/`. Restart Kiro IDE, or switch to the `kirograph` agent in Kiro CLI. It will now use KiroGraph tools automatically.
 
 Or using the short alias:
 
@@ -172,11 +179,11 @@ kg install
 └───────────────────────────────────────────┘
 ```
 
-Kiro hooks mark the index dirty on every file save or create, then flush on agent idle — batching changes efficiently with no overhead during active editing.
+Kiro hooks mark the index dirty on every file save or create, then flush on agent idle, batching changes efficiently with no overhead during active editing.
 
 ## Using with Kiro
 
-`kirograph install` sets up four things in your Kiro workspace — all coexist, so you can switch between IDE and CLI freely:
+`kirograph install` or `kirograph install --target kiro` sets up four things in your Kiro workspace (all coexist, so you can switch between IDE and CLI freely):
 
 ### MCP Server (`.kiro/settings/mcp.json`)
 
@@ -193,7 +200,8 @@ Registers the KiroGraph MCP server. Used by both the IDE and the CLI agent:
         "kirograph_callees", "kirograph_impact", "kirograph_node",
         "kirograph_status", "kirograph_files", "kirograph_dead_code",
         "kirograph_circular_deps", "kirograph_path", "kirograph_type_hierarchy",
-        "kirograph_architecture", "kirograph_coupling", "kirograph_package"
+        "kirograph_architecture", "kirograph_coupling", "kirograph_package",
+        "kirograph_hotspots", "kirograph_surprising", "kirograph_diff"
       ]
     }
   }
@@ -221,9 +229,9 @@ A custom agent for Kiro CLI that wires up the MCP server, inlines the steering i
 
 | Hook | Event | Action |
 |------|-------|--------|
-| `agentSpawn` | Agent starts | `kirograph sync-if-dirty --quiet` — catches edits made between sessions |
-| `userPromptSubmit` | Each prompt | `kirograph sync-if-dirty --quiet` — keeps graph fresh within a session |
-| `stop` | End of each turn | `kirograph sync-if-dirty --quiet` — deferred flush, mirrors IDE `agentStop` |
+| `agentSpawn` | Agent starts | `kirograph sync-if-dirty --quiet` (catches edits made between sessions) |
+| `userPromptSubmit` | Each prompt | `kirograph sync-if-dirty --quiet` (keeps graph fresh within a session) |
+| `stop` | End of each turn | `kirograph sync-if-dirty --quiet` (deferred flush, mirrors IDE `agentStop`) |
 
 Use it with:
 
@@ -243,13 +251,51 @@ Or swap to it inside an active session:
 
 Teaches the Kiro IDE to prefer graph tools over file scanning when `.kirograph/` exists. The CLI agent has the same instructions inlined directly in its `prompt` field.
 
+## Other Tools (Experimental)
+
+> **⚠️ Not fully tested, community-contributed.** The integrations below are outside the original scope of KiroGraph. They are provided as-is. Issues and PRs related to these targets are welcome, but there is no guarantee they will be supported or merged without active help from the contributor.
+
+KiroGraph can also be installed for other MCP-capable coding agents. All targets share the same `.kirograph/` data; if the project is already initialized, installing another target only writes that tool's integration files and reuses the existing graph.
+
+```bash
+kirograph install --target claude  # wire up Claude Code MCP + project memory
+kirograph install --target codex   # write Codex instructions and print MCP config
+```
+
+### Using with Claude Code
+
+```bash
+kirograph install --target claude
+```
+
+This writes:
+
+- `.mcp.json`: project-scoped MCP server config for Claude Code
+- `.kirograph/claude.md`: KiroGraph tool guidance
+- `CLAUDE.md`: an import of `.kirograph/claude.md`
+
+Claude Code prompts for project MCP approval the first time it sees `.mcp.json`.
+
+### Using with Codex
+
+```bash
+kirograph install --target codex
+```
+
+This writes:
+
+- `.kirograph/codex.md`: KiroGraph tool guidance
+- `AGENTS.md`: a generated KiroGraph instruction block
+
+Codex MCP configuration is user-scoped, so the installer prints the exact `codex mcp add ...` command and equivalent `~/.codex/config.toml` snippet instead of editing files outside the project.
+
 ## MCP Tools
 
-All tools are auto-approved and available to Kiro once installed.
+All tools are auto-approved in Kiro once installed. Other MCP clients can use the same tools after configuring their respective targets.
 
 ### `kirograph_context`
 
-Comprehensive context for a task or feature — often sufficient alone without additional tool calls.
+Comprehensive context for a task or feature, often sufficient alone without additional tool calls.
 
 | Parameter | Type | Default | Description |
 |-----------|------|---------|-------------|
@@ -267,7 +313,7 @@ Quick symbol search by name. Returns locations only, no code.
 | Parameter | Type | Default | Description |
 |-----------|------|---------|-------------|
 | `query` | string | required | Symbol name or partial name |
-| `kind` | string | — | Filter: `function`, `method`, `class`, `interface`, `type_alias`, `variable`, `route`, `component` |
+| `kind` | string | - | Filter: `function`, `method`, `class`, `interface`, `type_alias`, `variable`, `route`, `component` |
 | `limit` | number | 10 | Max results (1–100) |
 | `projectPath` | string | cwd | Project root path |
 
@@ -283,7 +329,7 @@ Find all functions/methods that call a specific symbol.
 | `limit` | number | 20 | Max results (1–100) |
 | `projectPath` | string | cwd | Project root path |
 
-**How it works:** BFS traversal of incoming `call` edges in the graph database. No vector engine involved.
+**How it works:** BFS traversal of incoming `call` edges in the graph database; no vector engine involved.
 
 ### `kirograph_callees`
 
@@ -295,7 +341,7 @@ Find all functions/methods that a specific symbol calls.
 | `limit` | number | 20 | Max results (1–100) |
 | `projectPath` | string | cwd | Project root path |
 
-**How it works:** BFS traversal of outgoing `call` edges in the graph database. No vector engine involved.
+**How it works:** BFS traversal of outgoing `call` edges in the graph database; no vector engine involved.
 
 ### `kirograph_impact`
 
@@ -307,7 +353,7 @@ Analyze what code would be affected by changing a symbol. Use before making chan
 | `depth` | number | 2 | Traversal depth |
 | `projectPath` | string | cwd | Project root path |
 
-**How it works:** BFS traversal of all incoming edges (`call`, `import`, `reference`, etc.) up to the specified depth. No vector engine involved.
+**How it works:** BFS traversal of all incoming edges (`call`, `import`, `reference`, etc.) up to the specified depth; no vector engine involved.
 
 ### `kirograph_node`
 
@@ -321,7 +367,7 @@ Get details about a specific symbol, optionally including source code.
 
 Returns: kind, name, qualified name, file location, signature, docstring, and optionally source code.
 
-**How it works:** Single row lookup by symbol name in the graph database. If `includeCode` is true, reads the relevant lines directly from the source file on disk. No vector engine involved.
+**How it works:** Single row lookup by symbol name in the graph database. If `includeCode` is true, reads the relevant lines directly from the source file on disk; no vector engine involved.
 
 ### `kirograph_type_hierarchy`
 
@@ -333,7 +379,7 @@ Traverse the type hierarchy of a class or interface.
 | `direction` | string | `both` | `up` (base types), `down` (derived types), `both` |
 | `projectPath` | string | cwd | Project root path |
 
-**How it works:** Recursive traversal of `extends` and `implements` edges in the graph database. No vector engine involved.
+**How it works:** Recursive traversal of `extends` and `implements` edges in the graph database; no vector engine involved.
 
 ### `kirograph_path`
 
@@ -345,7 +391,7 @@ Find the shortest path between two symbols in the dependency graph.
 | `to` | string | required | Target symbol name |
 | `projectPath` | string | cwd | Project root path |
 
-**How it works:** BFS shortest-path search across all edge types in the graph database. No vector engine involved.
+**How it works:** BFS shortest-path search across all edge types in the graph database; no vector engine involved.
 
 ### `kirograph_dead_code`
 
@@ -356,7 +402,7 @@ Find symbols with no incoming references (potential dead code). Only unexported
 | `limit` | number | 50 | Max results (1–100) |
 | `projectPath` | string | cwd | Project root path |
 
-**How it works:** Queries the graph database for nodes with zero incoming edges, filtered to non-exported symbols. No vector engine involved.
+**How it works:** Queries the graph database for nodes with zero incoming edges, filtered to non-exported symbols; no vector engine involved.
 
 ### `kirograph_circular_deps`
 
@@ -366,7 +412,7 @@ Find circular import dependencies in the codebase.
 |-----------|------|---------|-------------|
 | `projectPath` | string | cwd | Project root path |
 
-**How it works:** Tarjan's strongly connected components algorithm over `import` edges in the graph database. No vector engine involved.
+**How it works:** Tarjan's strongly connected components algorithm over `import` edges in the graph database; no vector engine involved.
 
 ### `kirograph_files`
 
@@ -374,14 +420,14 @@ List the indexed file structure with filtering and format options.
 
 | Parameter | Type | Default | Description |
 |-----------|------|---------|-------------|
-| `filterPath` | string | — | Filter by directory prefix (e.g., `src/`) |
-| `pattern` | string | — | Filter by glob pattern (e.g., `**/*.ts`) |
-| `maxDepth` | number | — | Limit tree depth |
+| `filterPath` | string | - | Filter by directory prefix (e.g., `src/`) |
+| `pattern` | string | - | Filter by glob pattern (e.g., `**/*.ts`) |
+| `maxDepth` | number | - | Limit tree depth |
 | `format` | string | `tree` | `tree`, `flat`, or `grouped` |
 | `includeMetadata` | boolean | true | Include language and symbol counts |
 | `projectPath` | string | cwd | Project root path |
 
-**How it works:** Reads file records from the graph database and builds a tree structure in memory. Filtering is applied before tree construction. No vector engine involved.
+**How it works:** Reads file records from the graph database and builds a tree structure in memory. Filtering is applied before tree construction; no vector engine involved.
 
 ### `kirograph_status`
 
@@ -411,10 +457,10 @@ Get coupling metrics for all packages or a specific one.
 
 | Parameter | Type | Default | Description |
 |-----------|------|---------|-------------|
-| `packageId` | string | — | Package ID (e.g. `pkg:npm:src/auth`). Omit for all packages. |
+| `packageId` | string | - | Package ID (e.g. `pkg:npm:src/auth`). Omit for all packages. |
 | `projectPath` | string | cwd | Project root path |
 
-Returns per-package: **Ca** (afferent — how many other packages depend on this one), **Ce** (efferent — how many packages this one depends on), and **instability** (`Ce / (Ca + Ce)`, 0 = maximally stable, 1 = maximally unstable). When `packageId` is given, also returns the full list of incoming and outgoing package dependencies.
+Returns per-package: **Ca** (afferent: how many other packages depend on this one), **Ce** (efferent: how many packages this one depends on), and **instability** (`Ce / (Ca + Ce)`, 0 = maximally stable, 1 = maximally unstable). When `packageId` is given, also returns the full list of incoming and outgoing package dependencies.
 
 ### `kirograph_package` *(requires `enableArchitecture: true`)*
 
@@ -447,7 +493,7 @@ Find non-obvious cross-file connections: direct edges between symbols in structu
 | `limit` | number | 20 | Max results (1–100) |
 | `projectPath` | string | cwd | Project root path |
 
-**How it works:** Queries all cross-file edges (excluding `contains` and `import`). Scores each by path distance between source and target files × edge-kind weight (`calls=1.0`, `references=0.8`, `type_of=0.7`, etc.). Returns the highest-scoring unique pairs — the ones that represent the most unexpected coupling in the codebase.
+**How it works:** Queries all cross-file edges (excluding `contains` and `import`). Scores each by path distance between source and target files × edge-kind weight (`calls=1.0`, `references=0.8`, `type_of=0.7`, etc.). Returns the highest-scoring unique pairs. the ones that represent the most unexpected coupling in the codebase.
 
 ### `kirograph_diff`
 
@@ -460,6 +506,31 @@ Compare the current graph state against a saved snapshot. Shows added/removed sy
 
 Use `kirograph snapshot save` (CLI) to save a snapshot before a refactor or PR. Run `kirograph_diff` after to see what changed structurally.
 
+### `kirograph_exec`
+
+Run a shell command and return token-optimized output. Automatically filters noise from git, test runners, linters, build tools, docker, and package managers.
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `command` | string | required | Shell command to execute |
+| `cwd` | string | project root | Working directory |
+| `level` | string | `normal` | Compression level: `normal`, `aggressive`, `ultra` |
+| `timeout` | number | 60 | Timeout in seconds |
+| `projectPath` | string | cwd | Project root path |
+
+**How it works:** Executes the command, detects the command family (git, test, lint, etc.), applies the appropriate filter strategy, and returns compressed output with a savings footer. Error output is always preserved. Does not require KiroGraph to be initialized, works standalone.
+
+### `kirograph_gain`
+
+Show token savings statistics from compressed command outputs.
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `period` | string | `session` | Time period: `session`, `today`, `week`, `all` |
+| `projectPath` | string | cwd | Project root path |
+
+Returns total commands, original/compressed token counts, savings percentage, breakdown by command family, and recent command history.
+
 ## CLI Reference
 
 ### Setup
@@ -468,8 +539,8 @@ Use `kirograph snapshot save` (CLI) to save a snapshot before a refactor or PR.
 kirograph install                 # Wire up MCP + hooks + steering in .kiro/
 kirograph init [path]             # Initialize .kirograph/ in a project
 kirograph init --index            # Initialize and index immediately
-kirograph uninit [path]           # Remove .kirograph/, hooks, and steering file
-kirograph uninit --force          # Skip confirmation prompt
+kirograph uninit [path]           # Prompts to remove integration files and .kirograph/ data
+kirograph uninit --force          # Remove everything without confirmation
 ```
 
 ### Indexing
@@ -526,7 +597,7 @@ Extracts symbol tokens from the task description (CamelCase, snake_case, SCREAMI
 
 ### Affected Tests
 
-Find test files that depend on changed source files — useful in CI or pre-commit hooks.
+Find test files that depend on changed source files, useful in CI or pre-commit hooks.
 
 ```bash
 kirograph affected src/utils.ts src/api.ts           # Pass files as arguments
@@ -561,7 +632,7 @@ fi
 
 Caveman mode compresses the agent's communication style, cutting token usage on responses without affecting tool calls or code output. Inspired by [caveman](https://github.com/JuliusBrussee/caveman) 🪨 by [JuliusBrussee](https://github.com/JuliusBrussee).
 
-**Why it's useful:** KiroGraph's graph tools return compact, structured data. The bottleneck in long coding sessions isn't the tool calls — it's the verbose prose the agent wraps around them. Caveman mode strips that overhead so you get the signal without the filler. The rules are injected at session start via the steering file (IDE) and the inline agent prompt (kiro-cli), so they're always in context with no extra tool calls.
+**Why it's useful:** KiroGraph's graph tools return compact, structured data. The bottleneck in long coding sessions isn't the tool calls; it is the verbose prose the agent wraps around them. Caveman mode strips that overhead so you get the signal without the filler. The rules are injected at session start via the steering file (IDE) and the inline agent prompt (kiro-cli), so they're always in context with no extra tool calls.
 
 Four levels:
 
@@ -582,10 +653,167 @@ kirograph caveman         # show current mode
 
 Set during `kirograph install` (interactive arrow-key menu) or any time after. Takes effect on the next agent session.
 
-Caveman mode never touches code blocks, file paths, URLs, or technical terms — only prose.
+Caveman mode never touches code blocks, file paths, URLs, or technical terms, only prose.
 
 **Auto-clarity exceptions:** the agent temporarily reverts to normal prose for security warnings, confirmations of irreversible actions (delete, overwrite, force-push), and multi-step sequences where fragment order could cause misunderstanding. Compressed style resumes immediately after.
 
+### Shell Compression (`kirograph_exec`)
+
+KiroGraph includes a built-in shell compression engine inspired by [rtk](https://github.com/rtk-ai/rtk). The `kirograph_exec` MCP tool runs shell commands and returns token-optimized output, saving 60-90% of tokens on verbose commands like git, test runners, linters, and build tools.
+
+**Why it's useful:** LLM context is expensive. A raw `git status` might be 2,000 tokens; compressed it's 200. A passing test suite might be 25,000 tokens of noise; compressed it's a single "PASSED: 42/42 tests" line. The compression engine knows how to extract the signal from each command family.
+
+Supported command families:
+
+| Family | Commands | Typical savings |
+|--------|----------|----------------|
+| Git | status, log, diff, push, pull, commit, add, fetch, branch, stash | 75-96% |
+| GitHub CLI | gh pr list/view, gh issue list, gh run list/view | 60-80% |
+| Test runners | jest, vitest, pytest, cargo test, go test, rspec, minitest, playwright | 80-90% |
+| Linters/build | eslint, tsc, ruff, clippy, cargo build, prettier, biome, golangci-lint, rubocop, next build | 70-85% |
+| File listings | ls, find, tree | 60-80% |
+| Search | grep, rg/ripgrep (grouped by file) | 60-80% |
+| Diff | diff file1 file2 (condensed context) | 50-70% |
+| Docker/k8s | docker ps, images, logs, compose ps, kubectl pods, logs, services | 70-80% |
+| Package managers | npm/pnpm install/list, pip list/install/outdated, bundle install/list, prisma generate | 75-92% |
+| AWS | sts, ec2, lambda, logs, cloudformation, dynamodb, iam, s3, ecs, sqs, sns | 60-88% |
+| Network | curl (strip progress/headers), wget (strip progress bars) | 50-70% |
+
+**Supported commands (full list):**
+
+```
+# Git
+kirograph exec git status                  # Compact status
+kirograph exec git log -n 10               # One-line commits
+kirograph exec git diff                    # Condensed diff
+kirograph exec git add .                   # → "ok"
+kirograph exec git commit -m "msg"         # → "ok abc1234"
+kirograph exec git push                    # → "ok main → origin/main"
+kirograph exec git pull                    # → "ok 3 files +10 -2"
+
+# GitHub CLI
+kirograph exec gh pr list                  # Compact PR listing
+kirograph exec gh pr view 42               # PR details + checks
+kirograph exec gh issue list               # Compact issue listing
+kirograph exec gh run list                 # Workflow run status
+
+# Test Runners
+kirograph exec jest                        # Failures only
+kirograph exec vitest run                  # Failures only
+kirograph exec playwright test             # E2E results (failures only)
+kirograph exec pytest                      # Python tests (-90%)
+kirograph exec go test ./...               # Go tests (-90%)
+kirograph exec cargo test                  # Cargo tests (-90%)
+kirograph exec rake test                   # Ruby minitest (-90%)
+kirograph exec rspec                       # RSpec tests (-60%+)
+
+# Build & Lint
+kirograph exec eslint .                    # Grouped by rule/file
+kirograph exec tsc --noEmit                # TypeScript errors grouped by file
+kirograph exec next build                  # Next.js build compact
+kirograph exec prettier --check .          # Files needing formatting
+kirograph exec cargo build                 # Cargo build (-80%)
+kirograph exec cargo clippy                # Cargo clippy (-80%)
+kirograph exec ruff check                  # Python linting (-80%)
+kirograph exec golangci-lint run           # Go linting (-85%)
+kirograph exec rubocop                     # Ruby linting (-60%+)
+kirograph exec biome check .               # Biome linting
+
+# Files & Search
+kirograph exec ls -la src/                 # Structured directory listing
+kirograph exec find . -name "*.ts"         # Grouped by directory
+kirograph exec tree                        # Truncated with summary
+kirograph exec grep -r "pattern" .         # Grouped search results
+kirograph exec rg "pattern"                # Grouped search results
+kirograph exec diff file1 file2            # Condensed diff
+
+# Package Managers
+kirograph exec npm install                 # → "ok +5 packages"
+kirograph exec npm list                    # Compact dependency tree
+kirograph exec pip list                    # Python packages
+kirograph exec pip install -r req.txt      # Strip progress bars
+kirograph exec bundle install              # Strip "Using" lines
+kirograph exec prisma generate             # Strip ASCII art
+
+# AWS
+kirograph exec aws sts get-caller-identity # One-line identity
+kirograph exec aws ec2 describe-instances  # Compact instance list
+kirograph exec aws lambda list-functions   # Name/runtime/memory
+kirograph exec aws logs get-log-events ... # Timestamped messages only
+kirograph exec aws cloudformation describe-stack-events ...  # Failures first
+kirograph exec aws dynamodb scan ...       # Unwraps type annotations
+kirograph exec aws iam list-roles          # Strips policy documents
+kirograph exec aws s3 ls s3://bucket       # Truncated listing
+
+# Containers
+kirograph exec docker ps                   # Compact container list
+kirograph exec docker images               # Compact image list
+kirograph exec docker logs container       # Deduplicated logs
+kirograph exec docker compose ps           # Compose services
+kirograph exec kubectl get pods            # Compact pod list
+kirograph exec kubectl logs pod            # Deduplicated logs
+kirograph exec kubectl get svc             # Compact service list
+
+# Network
+kirograph exec curl https://api.example.com/data  # Strip progress/headers
+kirograph exec wget https://example.com/file.zip  # Strip progress bars
+```
+
+Three compression levels:
+
+| Level | Style |
+|-------|-------|
+| `normal` | Balanced: removes noise, keeps structure *(default)* |
+| `aggressive` | More compact: groups by category, limits output |
+| `ultra` | Maximum compression: counts and summaries only |
+
+```bash
+kirograph compression normal     # balanced (default)
+kirograph compression aggressive # more compact
+kirograph compression ultra      # maximum compression
+kirograph compression off        # disable hook (tool still available)
+kirograph compression            # show current level
+```
+
+Set during `kirograph install` (interactive arrow-key menu) or any time after. When set to anything other than `off`, a `preToolUse` hook reminds the agent to use `kirograph_exec` for supported commands. The configured level is used as the default when the agent doesn't specify one explicitly.
+
+**Error preservation:** Failed commands always show full diagnostic output regardless of compression level. The engine detects error patterns and preserves detail when it matters.
+
+**Token analytics:**
+
+```bash
+kirograph gain               # summary stats
+kirograph gain --graph       # ASCII graph (last 30 days)
+kirograph gain --history     # recent command history
+kirograph gain --daily       # day-by-day breakdown
+kirograph gain --json        # JSON export
+```
+
+The `kirograph_gain` MCP tool exposes the same stats to the agent.
+
+### Savings Heuristics
+
+`kirograph gain` tracks two types of savings: compression (measured exactly) and graph tools (estimated via heuristics). For graph tools, the system estimates what the agent *would have spent* doing the same work without KiroGraph, based on typical agent behavior:
+
+| Tool | What the agent would do manually | Estimated naive cost |
+|------|----------------------------------|---------------------|
+| `kirograph_context` | Read 5-10 files to orient on a task | ~7,500-15,000 tokens |
+| `kirograph_search` | Run grep + read top matches | ~3,300 tokens |
+| `kirograph_callers` | Grep for symbol + read each calling file | ~8,300 tokens |
+| `kirograph_callees` | Read function body + grep for each call | ~3,900 tokens |
+| `kirograph_impact` | Recursive grep + read per depth level | ~6,900 × depth |
+| `kirograph_node` | Read the full file containing the symbol | ~1,500 tokens |
+| `kirograph_files` | Run `find` or `ls -R` | ~2,000 tokens |
+| `kirograph_path` | Trace connections manually (multiple grep + read) | ~7,700 tokens |
+| `kirograph_type_hierarchy` | Grep for extends/implements + read each file | ~5,400 tokens |
+| `kirograph_dead_code` | Not feasible manually (read every file) | 5× output, min 15,000 |
+| `kirograph_hotspots` | Not feasible manually (count edges for every symbol) | 5× output, min 15,000 |
+| `kirograph_architecture` | Not feasible manually | 4× output, min 7,500 |
+
+Constants used: 1,500 tokens per average source file (~200 lines), 800 tokens per grep result set, 2,000 tokens per directory listing. These are conservative estimates; in practice agents often read more files, retry failed searches, and explore dead ends.
+
+**Coexistence with Caveman Mode:** Compression and caveman mode are complementary, they compress different things. Caveman mode compresses the agent's *prose responses* (the text it writes around tool results); it never touches code or tool output. Shell compression compresses *shell command output* (the raw data coming back from shell commands); it never touches how the agent communicates. They stack: with both enabled, shell commands return 60-90% fewer tokens *and* the agent's explanations around those results are also shorter. Pick both independently during `kirograph install`. The "ultra + ultra" combo gives maximum token savings on both fronts.
+
 ### Architecture Analysis *(requires `enableArchitecture: true`)*
 
 Visualize the detected package graph, architectural layers, and package dependencies.
@@ -630,9 +858,9 @@ kirograph coupling --format json           # JSON output
 ```
 
 The table shows each package with:
-- **Ca** — afferent coupling: how many packages depend on this one (higher = more stable)
-- **Ce** — efferent coupling: how many packages this one depends on (higher = more unstable)
-- **Instability** — `Ce / (Ca + Ce)`, rendered as a color-coded bar: green (stable) → yellow (neutral) → red (unstable)
+- **Ca**: afferent coupling: how many packages depend on this one (higher = more stable)
+- **Ce**: efferent coupling: how many packages this one depends on (higher = more unstable)
+- **Instability** (`Ce / (Ca + Ce)`), rendered as a color-coded bar: green (stable) → yellow (neutral) → red (unstable)
 
 The `--package` detail view shows who depends on this package and what it depends on, with import counts for each relationship.
 
@@ -662,7 +890,7 @@ Score = path distance between files × edge-kind weight (`calls=1.0`, `reference
 
 ### Snapshots & Diff
 
-Save lightweight graph snapshots and compare them to track structural changes over time — useful before/after refactors, or in CI to audit what a PR added or removed.
+Save lightweight graph snapshots and compare them to track structural changes over time, useful before/after refactors, or in CI to audit what a PR added or removed.
 
 ```bash
 kirograph snapshot save [label]       # Save current graph state with optional label
@@ -674,13 +902,13 @@ kirograph snapshot diff --format full # Show full added/removed symbol lists
 kirograph snapshot diff --format json # JSON output
 ```
 
-Snapshots are stored in `.kirograph/snapshots/` as JSON and include all node IDs and edge tuples. The diff is computed as a set operation — O(n) regardless of codebase size.
+Snapshots are stored in `.kirograph/snapshots/` as JSON and include all node IDs and edge tuples. The diff is computed as a set operation, O(n) regardless of codebase size.
 
 The `kirograph_diff` MCP tool exposes the same capability to the agent: compare the current graph against the latest (or a named) snapshot without leaving the conversation.
 
 ### Dead Code
 
-Find unexported symbols with zero incoming references — candidates for removal.
+Find unexported symbols with zero incoming references, candidates for removal.
 
 ```bash
 kirograph dead-code [path]            # List dead code grouped by file
@@ -704,7 +932,7 @@ The command resolves symbol names using the same fuzzy search as `kirograph quer
 
 ### Graph Export
 
-Export the full graph as an interactive dashboard — three files served from a local directory, no server required, works offline.
+Export the full graph as an interactive dashboard. three files served from a local directory, no server required, works offline.
 
 ```bash
 kirograph export build [path]             # Generate .kirograph/export/{index.html,app.css,app.js}
@@ -721,10 +949,10 @@ Output lands in `.kirograph/export/` by default. Open `index.html` in any browse
 
 - **Color-coded nodes** by kind (class, function, method, component…) with size proportional to degree
 - **Directed edges** with kind labels; dashed lines for imports and references
-- **Click a node** to zoom in and inspect it — kind, file, line, degree, signature, and a copy button for the file reference
+- **Click a node** to zoom in and inspect it. kind, file, line, degree, signature, and a copy button for the file reference
 - **Click two nodes** to instantly find and highlight the shortest path between them, with detail cards for both endpoints
-- **History** — ‹ › navigation through previously inspected nodes
-- **Keyboard shortcuts** — `f` to fit the graph, `Esc` to exit focus or path mode
+- **History**: ‹ › navigation through previously inspected nodes
+- **Keyboard shortcuts**: `f` to fit the graph, `Esc` to exit focus or path mode
 
 #### Controls
 
@@ -746,9 +974,9 @@ Type to search by name, qualified name, or file path. Matching nodes are highlig
 
 #### Legend & filters
 
-- **Node kind filter** — Legend tab; click any kind to hide or show all nodes of that type
-- **Edge kind filter** — Legend tab; click any edge kind to hide or show edges of that type
-- **Degree slider** — Filters tab; hide nodes below N connections to surface the most-connected symbols
+- **Node kind filter**: Legend tab; click any kind to hide or show all nodes of that type
+- **Edge kind filter**: Legend tab; click any edge kind to hide or show edges of that type
+- **Degree slider**: Filters tab; hide nodes below N connections to surface the most-connected symbols
 
 #### Minimap
 
@@ -766,7 +994,7 @@ The 📊 Charts button opens a panel with three charts:
 |-------|--------------|
 | **Bar** | The 15 most-connected symbols |
 | **Donut** | How node kinds are distributed across the codebase |
-| **Line** | How many symbols have each connection count — reveals the overall connectivity shape of the graph |
+| **Line** | How many symbols have each connection count. reveals the overall connectivity shape of the graph |
 
 
 ### Dashboard
@@ -783,7 +1011,7 @@ kirograph dashboard stop [path]    # Stop the running engine server
 Reads `semanticEngine` from `.kirograph/config.json` and dispatches accordingly:
 
 - **qdrant**: Downloads the [Qdrant Web UI](https://github.com/qdrant/qdrant-web-ui) on first use (cached at `.kirograph/qdrant/dashboard/`), spawns the Qdrant server with `QDRANT__SERVICE__STATIC_CONTENT_DIR` set so the dashboard is served natively, and opens `http://127.0.0.1:<port>/dashboard` in your browser. If the server is already running with the dashboard, reconnects instead of restarting.
-- **typesense**: Downloads the [Typesense Dashboard](https://github.com/bfritscher/typesense-dashboard) static UI on first use (cached at `.kirograph/typesense/dashboard/`), starts the Typesense server if not already running, serves the dashboard locally via a Node HTTP server, and opens it in your browser. Press Ctrl+C to stop the dashboard server — the Typesense server keeps running as a background daemon.
+- **typesense**: Downloads the [Typesense Dashboard](https://github.com/bfritscher/typesense-dashboard) static UI on first use (cached at `.kirograph/typesense/dashboard/`), starts the Typesense server if not already running, serves the dashboard locally via a Node HTTP server, and opens it in your browser. Press Ctrl+C to stop the dashboard server. the Typesense server keeps running as a background daemon.
 
 Both servers run as persistent daemons. The state file (`.kirograph/qdrant-server.json` or `.kirograph/typesense-server.json`) tracks the PID and port for reconnection across `kg` commands.
 
@@ -816,10 +1044,11 @@ KiroGraph stores its config in `.kirograph/config.json`. You can edit it directl
 | `semanticEngine` | string | `cosine` | Search engine: `cosine`, `sqlite-vec`, `orama`, `pglite`, `lancedb`, `qdrant`, or `typesense` |
 | `useVecIndex` | boolean | `false` | Deprecated alias for `semanticEngine: "sqlite-vec"` |
 | `enableArchitecture` | boolean | `false` | Enable architecture analysis (package graph + layer detection, opt-in) |
-| `architectureLayers` | object | — | Custom layer definitions: `{ "layerName": ["glob/**"] }` |
+| `architectureLayers` | object | - | Custom layer definitions: `{ "layerName": ["glob/**"] }` |
 | `minLogLevel` | string | `warn` | Log level: `debug`, `info`, `warn`, `error` |
 | `fuzzyResolutionThreshold` | number | `0.5` | Name matching threshold for cross-file resolution (0.0–1.0) |
 | `cavemanMode` | string | `off` | Agent communication style: `off`, `lite`, `full`, `ultra` |
+| `shellCompressionLevel` | string | `normal` | Shell command compression level: `off`, `normal`, `aggressive`, `ultra` |
 
 Default exclude patterns: `node_modules/**`, `dist/**`, `build/**`, `.git/**`, `*.min.js`, `.kirograph/**`
 
@@ -833,7 +1062,7 @@ By default, KiroGraph uses exact name lookup and full-text search. Enable semant
 }
 ```
 
-This generates vector embeddings for all functions, methods, classes, interfaces, type aliases, components, and modules using a local embedding model (downloaded automatically to `~/.kirograph/models/` on first use). Embeddings are kept in sync automatically via Kiro hooks — on every file save, create, or delete.
+This generates vector embeddings for all functions, methods, classes, interfaces, type aliases, components, and modules using a local embedding model (downloaded automatically to `~/.kirograph/models/` on first use). Embeddings are kept in sync automatically via Kiro hooks. on every file save, create, or delete.
 
 Run `kirograph install` to be guided through model and engine selection interactively with arrow-key menus, or set the fields manually in `.kirograph/config.json`.
 
@@ -847,7 +1076,7 @@ Run `kirograph install` to be guided through model and engine selection interact
 | `onnx-community/embeddinggemma-300m-ONNX` | 768 | ~300MB | Google Gemma-based. Multilingual, 2048-token context window. |
 | `Xenova/all-MiniLM-L6-v2` | 384 | ~23MB | Lightweight, fast. Lower accuracy. |
 | `BAAI/bge-base-en-v1.5` | 768 | ~110MB | Strong general-purpose alternative to nomic. |
-| Custom | any | — | Any HuggingFace `feature-extraction` model. Provide ID + output dimension. |
+| Custom | any | - | Any HuggingFace `feature-extraction` model. Provide ID + output dimension. |
 
 The embedding dimension is stored in `embeddingDim` in `.kirograph/config.json` and used to initialise all vector engines correctly. Switching models requires a full re-index (`kirograph index --force`).
 
@@ -863,7 +1092,7 @@ Configure manually:
 
 #### Storage architecture
 
-Each engine owns its embedding store exclusively — there is no redundant write to the main graph database:
+Each engine owns its embedding store exclusively. there is no redundant write to the main graph database:
 
 | Engine | Graph store | Vector store |
 |--------|-------------|--------------|
@@ -881,7 +1110,7 @@ The graph store (`kirograph.db`) always holds nodes, edges, files, and all struc
 
 | Engine | Search type | Extra deps | Native? | Best for |
 |--------|-------------|------------|---------|----------|
-| `cosine` *(default)* | Exact cosine, linear scan | none | — | Small / medium projects, zero setup |
+| `cosine` *(default)* | Exact cosine, linear scan | none | - | Small / medium projects, zero setup |
 | `sqlite-vec` | ANN (approximate), sub-linear | `better-sqlite3`, `sqlite-vec` | yes | Large codebases, fast ANN search |
 | `orama` | Hybrid (full-text + vector) | `@orama/orama`, `@orama/plugin-data-persistence` | no (pure JS) | Best result quality, no native deps |
 | `pglite` | Hybrid (full-text + vector), exact | `@electric-sql/pglite` | no (pure WASM) | Exact results, no native deps, PostgreSQL semantics |
@@ -904,7 +1133,7 @@ In-process cosine similarity over all stored embeddings. No extra dependencies.
 
 #### sqlite-vec
 
-Approximate nearest-neighbour (ANN) index stored in `.kirograph/vec.db`. Sub-linear search time — ideal for large codebases with thousands of indexed symbols. The SQLite `vectors` table is not written to; `vec.db` is the sole embedding store.
+Approximate nearest-neighbour (ANN) index stored in `.kirograph/vec.db`. Sub-linear search time. ideal for large codebases with thousands of indexed symbols. The SQLite `vectors` table is not written to; `vec.db` is the sole embedding store.
 
 ```json
 {
@@ -921,7 +1150,7 @@ Requires two native dependencies (compiled C extensions). If not installed, fall
 
 #### orama
 
-Hybrid search powered by [Orama](https://github.com/oramasearch/orama) — combines full-text relevance and vector similarity in a **single query**, producing higher-quality results than running the two searches separately. The index is persisted to `.kirograph/orama.json` and is the sole embedding store. Pure JS, no native compilation required.
+Hybrid search powered by [Orama](https://github.com/oramasearch/orama). combines full-text relevance and vector similarity in a **single query**, producing higher-quality results than running the two searches separately. The index is persisted to `.kirograph/orama.json` and is the sole embedding store. Pure JS, no native compilation required.
 
 ```json
 {
@@ -938,7 +1167,7 @@ If not installed, falls back to `cosine`.
 
 #### pglite
 
-Hybrid search powered by [PGlite](https://github.com/electric-sql/pglite) — a WASM-compiled PostgreSQL with the [pgvector](https://github.com/pgvector/pgvector) extension. Combines **exact** nearest-neighbour vector search with full-text ranking (`ts_rank`) in a single SQL query. The database is persisted to `.kirograph/pglite/` using PostgreSQL's WAL-based storage and is the sole embedding store. Pure WASM — no native compilation required.
+Hybrid search powered by [PGlite](https://github.com/electric-sql/pglite), a WASM-compiled PostgreSQL with the [pgvector](https://github.com/pgvector/pgvector) extension. Combines **exact** nearest-neighbour vector search with full-text ranking (`ts_rank`) in a single SQL query. The database is persisted to `.kirograph/pglite/` using PostgreSQL's WAL-based storage and is the sole embedding store. Pure WASM, no native compilation required.
 
 ```json
 {
@@ -952,8 +1181,8 @@ npm install @electric-sql/pglite
 ```
 
 Key advantages:
-- **Exact** vector results (not approximate) — deterministic and reproducible
-- Native SQL `ON CONFLICT` upsert — no remove+insert workaround
+- **Exact** vector results (not approximate). deterministic and reproducible
+- Native SQL `ON CONFLICT` upsert, no remove+insert workaround
 - HNSW index (`vector_cosine_ops`) keeps search fast as the index grows
 - Single dependency, zero native binaries
 
@@ -961,7 +1190,7 @@ If not installed, falls back to `cosine`.
 
 #### LanceDB
 
-ANN vector search powered by [LanceDB](https://github.com/lancedb/lancedb) — stores embeddings in Apache Lance columnar format at `.kirograph/lancedb/`. Sub-linear search time using cosine distance. Pure JS, no native compilation required.
+ANN vector search powered by [LanceDB](https://github.com/lancedb/lancedb). stores embeddings in Apache Lance columnar format at `.kirograph/lancedb/`. Sub-linear search time using cosine distance. Pure JS, no native compilation required.
 
 ```json
 {
@@ -975,9 +1204,9 @@ npm install @lancedb/lancedb
 ```
 
 Key characteristics:
-- **Columnar storage** (Apache Lance format) — efficient for batch reads and writes
-- **ANN cosine search** — fast, sub-linear query time
-- Pure JS — no native binaries or WASM required
+- **Columnar storage** (Apache Lance format). efficient for batch reads and writes
+- **ANN cosine search**: fast, sub-linear query time
+- Pure JS, no native binaries or WASM required
 
 If not installed, falls back to `cosine`.
 
@@ -997,11 +1226,11 @@ npm install qdrant-local
 ```
 
 Key characteristics:
-- **HNSW index** — high-quality ANN search with Qdrant's native indexing
-- **Embedded binary** — no separate server setup; the process is spawned and managed automatically
-- **Persistent daemon** — the server stays running between `kg` commands; state tracked in `.kirograph/qdrant-server.json`
-- **Built-in dashboard** — run `kg dashboard start` to download the [Qdrant Web UI](https://github.com/qdrant/qdrant-web-ui) and open it (cached at `.kirograph/qdrant/dashboard/`, served via Qdrant's built-in static content feature)
-- **Async startup** — polls `/readyz` instead of blocking with a fixed sleep
+- **HNSW index**: high-quality ANN search with Qdrant's native indexing
+- **Embedded binary**: no separate server setup; the process is spawned and managed automatically
+- **Persistent daemon**: the server stays running between `kg` commands; state tracked in `.kirograph/qdrant-server.json`
+- **Built-in dashboard**: run `kg dashboard start` to download the [Qdrant Web UI](https://github.com/qdrant/qdrant-web-ui) and open it (cached at `.kirograph/qdrant/dashboard/`, served via Qdrant's built-in static content feature)
+- **Async startup**: polls `/readyz` instead of blocking with a fixed sleep
 - **Cosine distance** metric
 - Data persists across restarts in `.kirograph/qdrant/`
 
@@ -1030,11 +1259,11 @@ npm install typesense
 ```
 
 Key characteristics:
-- **HNSW index** — high-quality ANN search with Typesense's native indexing
-- **Auto-downloaded binary** — no manual server setup; the binary is fetched and cached at `~/.kirograph/bin/` on first run
-- **Persistent daemon** — the server stays running between `kg` commands; state tracked in `.kirograph/typesense-server.json`
-- **Local dashboard** — run `kg dashboard start` to open the built-in Typesense Dashboard UI (served locally, cached at `.kirograph/typesense/dashboard/`)
-- **Async startup** — polls `/health` instead of blocking with a fixed sleep
+- **HNSW index**: high-quality ANN search with Typesense's native indexing
+- **Auto-downloaded binary**: no manual server setup; the binary is fetched and cached at `~/.kirograph/bin/` on first run
+- **Persistent daemon**: the server stays running between `kg` commands; state tracked in `.kirograph/typesense-server.json`
+- **Local dashboard**: run `kg dashboard start` to open the built-in Typesense Dashboard UI (served locally, cached at `.kirograph/typesense/dashboard/`)
+- **Async startup**: polls `/health` instead of blocking with a fixed sleep
 - **Cosine distance** metric
 - Data persists across restarts in `.kirograph/typesense/`
 
@@ -1053,12 +1282,12 @@ When `enableArchitecture: true` is set, KiroGraph analyses the high-level struct
 
 #### What it detects
 
-**Packages** — logical groupings of files. Detected two ways:
+**Packages**: logical groupings of files. Detected two ways:
 
-1. **Manifest-based** — parsed from `package.json`, `go.mod`, `Cargo.toml`, `pyproject.toml`/`setup.py`/`setup.cfg`, `pom.xml`, `build.gradle`/`build.gradle.kts`, and `.csproj` files. Produces IDs like `pkg:npm:src/auth`.
-2. **Directory fallback** — for files not covered by any manifest, groups them by their nearest ancestor directory. Produces IDs like `pkg:dir:src/utils`.
+1. **Manifest-based**: parsed from `package.json`, `go.mod`, `Cargo.toml`, `pyproject.toml`/`setup.py`/`setup.cfg`, `pom.xml`, `build.gradle`/`build.gradle.kts`, and `.csproj` files. Produces IDs like `pkg:npm:src/auth`.
+2. **Directory fallback**: for files not covered by any manifest, groups them by their nearest ancestor directory. Produces IDs like `pkg:dir:src/utils`.
 
-**Layers** — architectural tiers detected from file paths using per-language glob patterns:
+**Layers**: architectural tiers detected from file paths using per-language glob patterns:
 
 | Layer | Examples |
 |-------|---------|
@@ -1070,12 +1299,12 @@ When `enableArchitecture: true` is set, KiroGraph analyses the high-level struct
 
 Layer detection is per-language (TypeScript/JS, Python, Go, Java, Ruby, Rust, C#) with framework-specific patterns where applicable (Django, Rails, Spring MVC, ASP.NET, etc.). Custom layer overrides are supported via `architectureLayers` in config.
 
-**Package dependencies** — rolled up from existing `imports` edges in the graph. No re-parsing required.
+**Package dependencies**: rolled up from existing `imports` edges in the graph. No re-parsing required.
 
-**Coupling metrics** — computed per package:
-- **Ca** (afferent) — how many other packages depend on this one
-- **Ce** (efferent) — how many packages this one depends on
-- **Instability** — `Ce / (Ca + Ce)`: 0 = maximally stable (everyone depends on it, it depends on nothing), 1 = maximally unstable (depends on everything, nobody depends on it)
+**Coupling metrics**: computed per package:
+- **Ca** (afferent). how many other packages depend on this one
+- **Ce** (efferent). how many packages this one depends on
+- **Instability** (`Ce / (Ca + Ce)`): 0 = maximally stable (everyone depends on it, it depends on nothing), 1 = maximally unstable (depends on everything, nobody depends on it)
 
 #### Custom layer definitions
 
@@ -1215,10 +1444,20 @@ Ansible
 
 Detected frameworks are stored in config and used to improve symbol extraction and resolution.
 
+## Credits
+
+KiroGraph is inspired by [CodeGraph](https://github.com/colbymchenry/codegraph) by [Colby McHenry](https://www.linkedin.com/in/colby-mchenry/). the original concept of building a semantic code graph for AI coding agents comes from his work.
+
+### Contributors
+
+- [Alessandro Franceschi](https://www.linkedin.com/in/alessandrofranceschi/). Claude Code and Codex integration, Elixir/Phoenix language and framework support.
+- [Mauro Argo](https://www.linkedin.com/in/argomauro/). original idea for the architecture layer analysis feature.
+
 ## Requirements
 
 - Node.js >= 18
-- Kiro IDE
+- Kiro IDE (fully supported)
+- Other MCP-capable tools (experimental. see [Other Tools](#other-tools-experimental))
 
 ## License