# Feature documentation
[Repository README](../../README.md)
Comprehensive user-facing feature docs for `magi-code`. README stays as overview; these files hold detailed behavior, examples, contracts, and troubleshooting.
## Feature map
| [Installation and updates](installation-and-updates.md) | Requirements, quick start, Cargo/source installs, GitHub tags, manual versioned updates, aliases. |
| [Configuration](configuration.md) | `MC_HOME`, `settings.json`, precedence, context settings, session titles, autocomplete, env/auth boundaries. |
| [Provider authentication](provider-authentication.md) | Custom OpenAI-compatible providers and OpenAI Codex OAuth login/logout. |
| [Prompt-cache audit scripts](prompt-cache-audit-scripts.md) | Maintainer static audit and opt-in live cold/warm checker for provider prompt-cache readiness. |
| [CPU hotspot profiling](cpu-hotspot-profiling.md) | Maintainer credential-free CPU profiling workflow, baseline tables, hotspot reports, and before/after comparison method. |
| [Memory usage profiling](memory-usage-profiling.md) | Maintainer credential-free peak RSS profiling workflow reusing CPU harness scenarios, artifact layout, and comparison method. |
| [Memory benchmark history](memory-benchmark-history.md) | Maintainer peak RSS run history, update rules, missing-field handling, and regression trigger guidance. |
| [fffind and content-search benchmarks](fff-benchmarks.md) | Maintainer credential-free FFF path/content benchmark harnesses, latest results, shipped `ffgrep` backend status, and interpretation rules. |
| [Eval harness and skill benches](eval-harness-and-skill-benches.md) | Maintainer prompt-fixture dump, tool-choice evals, native skill benches, candidate gate, SkillOpt bridge, and skill-specific scoring contracts. |
| [Duplicate detection](duplicate-detection.md) | Maintainer `cargo-dupes` and `jscpd` config, commands, thresholds, metrics, and duplicate cleanup policy. |
| [Commands and interfaces](commands-and-interfaces.md) | Startup flags, print mode, classic shell commands, Mission Control command surfaces. |
| [Mission Control TUI](mission-control-tui.md) | TUI launch constraints, transcript/activity behavior, modals, autocomplete, controls. |
| [Sessions, context, and cache](sessions-context-cache.md) | JSONL sessions, structured replay, context budget failure, local cache keys/catalogs. |
| [Instructions, prompts, skills, and primary agents](instructions-prompts-skills-and-agents.md) | AGENTS.md precedence, prompt overrides, skill roots, primary-agent profiles. |
| [Prompt construction](prompt-construction.md) | System prompt assembly, delimiters, dynamic context files, tool/skill/subagent injection, provider request mapping. |
| [Tools and safety model](tools-and-safety.md) | Built-in and MCP tool contracts, visibility, path/symlink boundaries, search-tool credential model. |
| [MCP stdio tools](mcp-stdio-tools.md) | `mcp_servers` settings, `mcp__server__tool` names, CLI diagnostics, stdio-only Phase 1 limits. |
| [LSP diagnostics](lsp-diagnostics.md) | Opt-in language-server diagnostics injection, `diagnostics` and `references` tools, caps, degradation, and privacy boundaries. |
| [Tool output compression](tool-output-compression.md) | Opt-in provider-visible curated summaries for recognized bash outputs while raw local/session results stay preserved. |
| [Tool-call bash hooks](tool-call-hooks.md) | Before/after hook settings, payloads, failure policies, script examples, privacy. |
| [Parallel subagents](parallel-subagents.md) | Delegation limits, concurrency, inherited context, identity profiles. |
| [Troubleshooting](troubleshooting.md) | Build/launch, auth/provider, session/context/skill fixes. |
| [Security notes](security.md) | Credential handling, package/update credential boundary, auth-file permissions, shell limits. |
## Maintainer notes
- Keep feature docs user-facing. Governance stays in [PRDs](../prd/) and [ADRs](../adr/).
- Maintainer duplicate detection workflow: [Duplicate detection](duplicate-detection.md).
- Do not add real API keys, OAuth tokens, refresh tokens, account ids, bearer headers, auth codes, or credential output to examples.
- When behavior changes, update README overview plus the relevant feature doc in the same change.