longbridge · endless-bot · Jul 3, 2026
diff --git a/CLAUDE.md b/CLAUDE.md
@@ -178,3 +178,24 @@ Keep SKILL.md under ~200 lines. Push detail (long field dictionaries, multi-page
 ## License
 
 MIT. See [LICENSE](./LICENSE).
+
+## Skill attribution protocol
+
+Every skill MUST set `LONGBRIDGE_SOURCE_SKILL=<skill-slug>` as an env-var prefix before each `longbridge` CLI call so the backend can attribute the request to this skill.
+
+**Prompt layer (all skills)**: each SKILL.md `## CLI` section starts with:
+
+```
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=<skill-slug>` so the backend can attribute
+> the request to this skill.
+```
+
+**Script layer (`longbridge-earnings`)**: `scripts/collect.py` passes `env={"LONGBRIDGE_SOURCE_SKILL": "longbridge-earnings", **os.environ}` to every `subprocess.run()` call.
+
+**Protocol details**:
+- Env var name: `LONGBRIDGE_SOURCE_SKILL`
+- Value format: `^[a-z0-9]+(-[a-z0-9]+)*$` — matches the skill's `name:` frontmatter field exactly
+- HTTP header (RFC, CLI-side): `X-Lb-Skill: <skill-slug>` — injected by the CLI binary when the env var is set (external dependency)
+- Precision: best-effort; CLI ignores the env var until the binary implements RFC §3.2
+- MCP path attribution: pending server-side support (RFC §3.3)
diff --git a/docs/architecture.md b/docs/architecture.md
@@ -92,8 +92,8 @@ The LLM reads `longbridge` stderr directly and translates the situation into the
    3. Response-language directive tells the LLM:
         user typed zh-Hant → reply in zh-Hant
                             ↓
-     longbridge quote NVDA.US --format json   (LLM may also call `static`
-                            ↓                   and `calc-index` and merge by symbol)
+     LONGBRIDGE_SOURCE_SKILL=longbridge-quote longbridge quote NVDA.US --format json
+                            ↓
        longbridge CLI returns a JSON array (English fields)
                             ↓
    4. LLM uses zh-Hant + the multilingual tables in §3/§4 to compose:
@@ -120,7 +120,7 @@ Every `SKILL.md` tells the LLM the same rule: **default to CLI; fall back to MCP
 ### Default rule
 
 ```
-1. LLM calls `longbridge <subcommand> --format json` directly.
+1. LLM calls `LONGBRIDGE_SOURCE_SKILL=<skill-slug> longbridge <subcommand> --format json` directly.
 2. Shell returns `command not found` → fall back to MCP.
 3. Otherwise parse the JSON output (or surface stderr verbatim if non-zero exit).
 4. MCP also unconfigured → tell the user: install longbridge-terminal or run `claude mcp add ...`.

diff --git a/skills/longbridge-content/SKILL.md b/skills/longbridge-content/SKILL.md
@@ -39,6 +39,11 @@ Trigger when user asks about: latest news for a stock, company announcements / r
 
 ## CLI Commands
 
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-content` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-content longbridge news TSLA.US --format json`
+
 Run `longbridge <cmd> --help` for current flags and output fields.
 
 ### `news` — latest news articles for a symbol; fetch full article content
@@ -65,6 +70,8 @@ All commands: Public — no login required.
 
 Use MCP server if CLI unavailable. Discover tools at runtime.
 
+> **MCP attribution**: MCP 路径的 skill 归因需服务端支持（RFC §3.3），当前为 best-effort。
+
 ## Related skills
 
 | User wants | Use |

diff --git a/skills/longbridge-derivatives/SKILL.md b/skills/longbridge-derivatives/SKILL.md
@@ -40,6 +40,11 @@ Trigger when user asks about: options quotes, option chains, Greeks (Delta/Gamma
 
 ## CLI Commands
 
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-derivatives` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-derivatives longbridge option chain TSLA.US --format json`
+
 ### `option` — option quotes, option chain, option volume statistics
 
 Run `longbridge option --help` for subcommands (quote / chain / volume).

diff --git a/skills/longbridge-earnings/SKILL.md b/skills/longbridge-earnings/SKILL.md
@@ -86,6 +86,12 @@ Read [references/full-report.md](references/full-report.md) and follow it. In sh
 
 ## Fallbacks
 
+> **Skill attribution**: before every direct `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-earnings` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-earnings longbridge financial-report snapshot TSLA.US --format json`
+> Note: `scripts/collect.py` already sets this env var automatically.
+
 - **Partial N/A sections**: the digest marks failed sources as `N/A (reason)`. Work with what succeeded; fetch a missing critical source directly (`longbridge <cmd> <SYMBOL> --format json`), checking `--help` only when a command errors.
 - **No Python (script-less path)**: issue the CLI calls yourself — in PARALLEL (multiple tool calls in one message), never sequentially, and keep raw output small: use `--format json` everywhere, `kline ... --count 30`, `news ... --count 10`, and SKIP the full income statement (`financial-report --kind IS` is ~100KB raw) — take revenue/NI/EPS trends from `consensus` (it carries ~6 periods of estimate + actual) and margins from `financial-report snapshot`.
 - **HK symbols**: leading zeros are stripped automatically (`09988.HK` → `9988.HK`); do the same when calling the CLI directly.

diff --git a/skills/longbridge-earnings/scripts/collect.py b/skills/longbridge-earnings/scripts/collect.py
@@ -19,6 +19,7 @@
 """
 
 import json
+import os
 import re
 import shutil
 import subprocess
@@ -40,10 +41,12 @@ def die(msg, code):
 
 def fetch(out_dir, name, args):
     """Run one CLI call; save raw JSON to <name>.json or error to <name>.err."""
+    env = {"LONGBRIDGE_SOURCE_SKILL": "longbridge-earnings", **os.environ}
     try:
         proc = subprocess.run(
             ["longbridge", *args, "--format", "json"],
             capture_output=True, text=True, encoding="utf-8", timeout=TIMEOUT,
+            env=env,
         )
         ok = proc.returncode == 0 and proc.stdout.strip()
     except (subprocess.TimeoutExpired, OSError) as e:

diff --git a/skills/longbridge-fundamentals/SKILL.md b/skills/longbridge-fundamentals/SKILL.md
@@ -55,6 +55,11 @@ Trigger when user asks about: financial statements (income/balance sheet/cash fl
 
 ## CLI Commands
 
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-fundamentals` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-fundamentals longbridge financial-report TSLA.US --format json`
+
 Run `longbridge <cmd> --help` for current flags and output fields.
 
 ### `financial-report` — income statement, balance sheet, cash flow
@@ -107,6 +112,8 @@ All commands: Public — no login required.
 
 Use MCP server if CLI unavailable. Discover tools at runtime.
 
+> **MCP attribution**: MCP 路径的 skill 归因需服务端支持（RFC §3.3），当前为 best-effort。
+
 ## Related skills
 
 | User wants | Use |

diff --git a/skills/longbridge-intel/SKILL.md b/skills/longbridge-intel/SKILL.md
@@ -51,6 +51,11 @@ Trigger when user asks about: strategy screening, stock popularity rankings, top
 
 ## CLI Commands
 
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-intel` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-intel longbridge screener --format json`
+
 Run `longbridge <cmd> --help` for current flags and output fields.
 
 ### `screener` — strategy screener: browse strategies, run filters
@@ -112,6 +117,8 @@ All CLI commands: Public — no login required.
 
 Use MCP server if CLI unavailable. Discover tools at runtime.
 
+> **MCP attribution**: MCP 路径的 skill 归因需服务端支持（RFC §3.3），当前为 best-effort。
+
 ## Related skills
 
 | User wants | Use |

diff --git a/skills/longbridge-market-data/SKILL.md b/skills/longbridge-market-data/SKILL.md
@@ -52,6 +52,11 @@ Trigger when the user asks about: stock price / quote, K-line / candlestick char
 
 ## CLI Commands
 
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-market-data` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-market-data longbridge quote TSLA.US --format json`
+
 Run `longbridge --help` to list all subcommands. Run `longbridge <cmd> --help` for flags.
 
 ### `quote` — real-time quote for one or more symbols

diff --git a/skills/longbridge-portfolio/SKILL.md b/skills/longbridge-portfolio/SKILL.md
@@ -53,6 +53,11 @@ Trigger when user asks about: account assets / net value, stock or fund position
 
 ## CLI Commands
 
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-portfolio` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-portfolio longbridge portfolio --format json`
+
 Run `longbridge <cmd> --help` for current flags and output fields.
 
 ### `assets` — account net assets, cash, buying power, margin breakdown
@@ -112,6 +117,8 @@ Identify unrealised losses, suggest substitutes, track 30-day wash-sale window.
 
 Use MCP server if CLI unavailable. Discover tools at runtime.
 
+> **MCP attribution**: MCP 路径的 skill 归因需服务端支持（RFC §3.3），当前为 best-effort。
+
 ## Related skills
 
 | User wants | Use |

diff --git a/skills/longbridge-quant/SKILL.md b/skills/longbridge-quant/SKILL.md
@@ -47,6 +47,11 @@ Trigger when user asks about: quantitative indicator scripts (running against K-
 
 ## CLI: quant
 
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-quant` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-quant longbridge quant --help`
+
 The `quant` command runs user-defined indicator scripts against K-line data.
 
 ```bash
@@ -109,6 +114,8 @@ Rolling walk-forward Random Forest / Gradient Boosting, feature engineering, sig
 
 Use MCP server for kline data if CLI unavailable. Discover tools at runtime.
 
+> **MCP attribution**: MCP 路径的 skill 归因需服务端支持（RFC §3.3），当前为 best-effort。
+
 ## Related skills
 
 | User wants | Use |

diff --git a/skills/longbridge-research/SKILL.md b/skills/longbridge-research/SKILL.md
@@ -76,6 +76,11 @@ Trigger when user asks about: analyst ratings / price targets, EPS/revenue conse
 
 ## CLI Commands
 
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-research` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-research longbridge institution-rating TSLA.US --format json`
+
 Run `longbridge <cmd> --help` for current flags and output fields.
 
 ### `institution-rating` — buy/hold/sell distribution and recent rating events
@@ -142,6 +147,8 @@ All CLI commands: Public — no login required.
 
 Use MCP server if CLI unavailable. Discover tools at runtime.
 
+> **MCP attribution**: MCP 路径的 skill 归因需服务端支持（RFC §3.3），当前为 best-effort。
+
 ## Related skills
 
 | User wants | Use |

diff --git a/skills/longbridge-technical/SKILL.md b/skills/longbridge-technical/SKILL.md
@@ -29,6 +29,11 @@ Trigger when user asks about: candlestick patterns, Ichimoku cloud, technical in
 
 ## Data dependency
 
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-technical` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-technical longbridge kline TSLA.US --period day --count 200 --format json`
+
 ⚠️ All frameworks in this skill require OHLCV historical data. **Before running any analysis, fetch K-line data:**
 
 ```bash

diff --git a/skills/longbridge-value-investing/SKILL.md b/skills/longbridge-value-investing/SKILL.md
@@ -54,6 +54,15 @@ Hard quant filter (ROE ≥ 15%, debt ≤ 50%, FCF positive, gross margin ≥ 30%
 
 All frameworks: Public — no login required.
 
+## CLI
+
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-value-investing` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-value-investing longbridge financial-report TSLA.US --format json`
+
+Run `longbridge --help` to discover available subcommands; `longbridge <cmd> --help` for flags.
+
 ## Error handling
 
 | Situation | Response |

diff --git a/skills/longbridge-watchlist/SKILL.md b/skills/longbridge-watchlist/SKILL.md
@@ -37,6 +37,11 @@ Trigger when user asks about: viewing watchlist groups, adding or removing symbo
 
 ## CLI Commands
 
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-watchlist` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge-watchlist longbridge watchlist --format json`
+
 Run `longbridge <cmd> --help` for current flags and output fields.
 
 ### `watchlist` — list groups; create / rename / delete groups; add / remove symbols 🔐 ⚠️ mutating

diff --git a/skills/longbridge/SKILL.md b/skills/longbridge/SKILL.md
@@ -21,6 +21,11 @@ For setup and authentication details, see [references/setup.md](references/setup
 
 ## Investment Analysis Workflow
 
+> **Skill attribution**: before every `longbridge` CLI call, prefix it with
+> `LONGBRIDGE_SOURCE_SKILL=longbridge` so the backend can attribute
+> the request to this skill. Example:
+> `LONGBRIDGE_SOURCE_SKILL=longbridge longbridge quote TSLA.US --format json`
+
 When the user asks about stock performance, portfolio advice, or market analysis:
 
 1. **Get live data** via CLI — quotes, positions, K-line history, intraday