Skip to content
Skills Library
AI & agents · Learning & explanation

flightdeck

/flightdeck

Explains what the Flightdeck plugin records about this Claude Code session, where it shows up, which environment variables control capture and enforcement, and how MCP policy decisions reach the session. Use when the user asks what Flightdeck sees, why a tool call was refused by an MCP policy, or how to turn capture off.

Repository
flightdeck ↗
Licence
Apache-2.0
Last repository change
27.10.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
1
Scripts
0
HTML comments
0
URLs in SKILL.md
2
Deprecation

No signals detected. Repository changed on 27.10.2026, without references to retired models or APIs.

Licence

Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.

What it does and when to use it

Explains what the Flightdeck plugin records about this Claude Code session, where it shows up, which environment variables control capture and enforcement, and how MCP policy decisions reach the session. Use when the user asks what Flightdeck sees, why a tool call was refused by an MCP policy, or how to turn capture off.

Original author description

Explains what the Flightdeck plugin records about this Claude Code session, where it shows up, which environment variables control capture and enforcement, and how MCP policy decisions reach the session. Use when the user asks what Flightdeck sees, why a tool call was refused by an MCP policy, or how to turn capture off.

How it works

  1. What the plugin records
  2. Where the session shows up
  3. MCP policy enforcement
  4. Environment variables
  5. Failure behaviour
  6. Privacy
Original SKILL.md headings · 104 lines, 836 words
SKILL.md preview
---
name: flightdeck
description: Explains what the Flightdeck plugin records about this Claude Code session, where it shows up, which environment variables control capture and enforcement, and how MCP policy decisions reach the session. Use when the user asks what Flightdeck sees, why a tool call was refused by an MCP policy, or how to turn capture off.
---

# Flightdeck

Flightdeck is a self-hosted control plane for AI agents. With this plugin
loaded, the current Claude Code session streams lifecycle, tool and LLM
events to the Flightdeck server configured for this machine, and the
session appears in the Fleet view next to production agents.

## What the plugin records

Every hook Claude Code exposes runs the same script, so the session is
reported from start to end:

- Session start and end, including sub-agent sessions spawned through the
  Agent tool (reported under their own session id, linked to the parent).
- Every tool call by name, successful or failed, with Claude Code's
  measured duration and `tool_use_id`, a structured error on failure
  (its text only while both capture flags are on; `<redacted>`
  otherwise), and a sanitised whitelist of arguments (`file_path`,
  `command`, `query`, `pattern`, `prompt`) while
  `FLIGHTDECK_CAPTURE_TOOL_INPUTS` is on. MCP tool calls carry their full
  arguments because the whitelist would empty them, plus the server
  identity Claude Code reports (`mcp_server`) when it sends one.
- Every LLM turn read from the session transcript: model, input and output
  tokens, cache tokens (with their one-hour share), thinking tokens,
  service tier, data residency, web search counts, stop reason, latency,
  and the prompt and response bodies while `FLIGHTDECK_CAPTURE_PROMPTS`
  is on. Each turn also carries the session's billing mode: a Claude
  subscription login is `subscription` (not counted as spend), an API
  key or a Bedrock / Vertex AI / Foundry session is `metered`.
- Context compaction and the MCP servers declared in `.mcp.json`.
- MCP policy decisions: when a server is warned or blocked by the
  Flightdeck MCP policy for this flavor, the session emits a
  `policy_mcp_warn` or `policy_mcp_block` event.

Prompt capture is on by default for this plugin because you are observing
your own session on your own machine. The Python sensor used for
production agents defaults the other way.

## Where the session shows up

Open the dashboard served by `FLIGHTDECK_SERVER` (default
`http://localhost:4000`). The session is stamped with:

- flavor `claude-code`
- agent type `coding`
- client type `claude_code`
- a context block with the native Claude Code session id, the Claude
  Code version from the transcript, how the session started (known
  values include `startup`, `resume`, `clear` and `compact`), the
  entrypoint (known values include `cli` and `sdk-cli`), whether a human
  is attending it, and cloud / CI markers when present

Filter the Fleet view by flavor `claude-code` to see only Claude Code
sessions. The session drawer shows tool calls, LLM turns, token totals and
the Prompts tab (empty with a clear notice when capture is off).

## MCP policy enforcement

Before each `mcp__<server>__<tool>` call the plugin evaluates the MCP
policy fetched at session start for flavor `claude-code`:

- `allow` entries let the call through silently.
- `deny` entries with `block` enforcement refuse the call. The tool result
  reads `MCP policy blocked tool call to <server> (<scope>, <path>)`.
- A server unknown to an allowlist-mode policy triggers Claude Code's own
  permission prompt. Approving it once is remembered on this machine.

Operators change the policy in the dashboard under MCP policies. The
plugin re-reads it at the next session start.

## Environment variables

| Variable | Default | Effect |
| --- | --- | --- |
| `FLIGHTDECK_SERVER` | `http://localhost:4000` | Control plane base URL; events go to `/ingest`, policies come from `/api`. |
| `FLIGHTDECK_TOKEN` | `tok_dev` | Access token. The default is the development seed token and prints a warning; set a real token outside local development. |
| `FLIGHTDECK_CAPTURE_PROMPTS` | `true` | `false` strips prompt, response and tool-result bodies while keeping metadata and token counts. |
| `FLIGHTDECK_CAPTURE_TOOL_INPUTS` | `true` | `false` strips the sanitised tool arguments. |
| `FLIGHTDECK_BILLING_MODE` | detected | `metered`, `subscription` or `local`; overrides the billing mode the plugin detects from `claude auth status` and the cloud-provider variables (set `metered` when a subscription runs on extra usage). |
| `FLIGHTDECK_QUIET` | unset | `1` silences the default-token warning. |
| `CLAUDE_SESSION_ID` | unset | Forces the session id reported to Flightdeck. |
| `CLAUDE_CODE_SESSION_ID` | set by Claude Code | Claude Code's own session id; reported as `context.claude_code_session_id` and used to key the per-session marker when a hook payload has no `session_id`. |

## Failure behaviour

The plugin never blocks Claude Code. If the server is unreachable it
prints one line to stderr per failed request and exits successfully, and
the next hook tries again. The `session_end` is posted by a detached
worker after the hook returns (Claude Code allows SessionEnd hooks 1.5
seconds); its failed-request lines go to
`$TMPDIR/flightdeck-plugin/session-end-worker.log` (Windows:
`%TEMP%\flightdeck-plugin\session-end-worker.log`), an owner-only file
truncated past 256 KiB. Policy evaluation fails open: with no cached
policy, tool calls proceed as Claude Code would normally decide, and MCP
servers bundled with other plugins are reported but not gated.

## Privacy

Everything the plugin captures goes only to the Flightdeck server you
configured, never to a third party. Turn either capture flag off to keep
content bodies on your machine; the session, tool names and token counts
still ship so the dashboard stays useful.

Versions

The database has no other skills named flightdeck.

Database snapshot: 13.11.2026 · package version 89001fee66b3 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100mcp-client-configuration-claude-desktop
    Use this skill when registering, configuring environment variables for, and troubleshooting MCP servers in Claude Desktop configuration files.
  • 100cc-inspect
    Claude Code runtime environment inspector. Comprehensive diagnostic tool that reports all Claude Code environment variables, session identity, config directory, plugin paths, system toolchain versions, Rune runtime state, and platform details. Use when debugging environment…
  • 100furl
    How the Furl context-compression plugin works in Claude Code and Codex — the furl_compress / furl_retrieve / furl_stats / furl_purge / furl_search / furl_list MCP tools, the <<ccr:HASH>> retrieval flow, and Claude Code's optional automatic hooks. Use when the user asks what Furl…
  • 100session-capture
    Create a normalized session capture file preserving what happened in a session. Use at the end of a work session, or when the user says "capture session" or "save session". Records session goal, work performed, decisions, blockers, and verification results to .untracked/session-captures/.
  • 100auto-capture-claude-code
    Claude Code adapter for the auto-capture skill. Extends auto-capture with automatic session-end hooks that capture transcripts to Open Brain without manual intervention. Use this when you want every meaningful Claude Code session to be preserved automatically — not just the ones…
  • 100create-plugin
    This skill should be used when the user asks to "create a plugin", "make a GitHub plugin", "turn my repo into a plugin", "create a Claude Code plugin", "create a Codex plugin", "publish my skills as a plugin", or "set up a plugin marketplace". Scaffolds the complete file…

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.