Refine market scout, pitch rendering, and signal sweep skills

.github/workflows/insider-scan.yml

@@ -2,8 +2,7 @@ name: Daily Insider Scan
 
 on:
   schedule:
-    # 7:00 AM ET on weekdays (11:00 UTC Nov-Mar, 12:00 UTC Mar-Nov)
-    # Using 12:00 UTC to cover EDT; during EST this runs at 7 AM still fine
+    # 12:00 UTC on weekdays (7 AM EST / 8 AM EDT).
     - cron: "0 12 * * 1-5"
   workflow_dispatch:
     inputs:
@@ -27,8 +26,14 @@ jobs:
         with:
           python-version: "3.12"
 
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+
       - name: Install dependencies
-        run: pip install -r skills/signal-sweep/requirements.txt
+        working-directory: skills/signal-sweep
+        run: uv sync --locked
 
       - name: Run insider scan
         env:
@@ -40,16 +45,10 @@ jobs:
           LOOKBACK="${{ github.event.inputs.lookback || '5' }}"
           ZSCORE="${{ github.event.inputs.zscore || '1.5' }}"
 
-          WEBHOOK_ARG=""
-          if [ -n "$DISCORD_WEBHOOK_URL" ]; then
-            WEBHOOK_ARG="--webhook $DISCORD_WEBHOOK_URL"
-          fi
-
-          python scripts/scan_insiders.py \
+          uv run python scripts/scan_insiders.py \
             --date "$DATE" \
             --lookback "$LOOKBACK" \
-            --zscore "$ZSCORE" \
-            $WEBHOOK_ARG
+            --zscore "$ZSCORE"
 
       - name: Upload scan results
         if: always()

.markdownlint-cli2.jsonc

@@ -11,7 +11,9 @@
   },
   "ignores": [
     ".venv",
+    "**/.venv/**",
     "node_modules",
+    "**/node_modules/**",
     ".git",
     "**/sec-cache/**",
     "**/signal-sweep-cache/**",

skills/README.md

@@ -54,62 +54,8 @@ the skill that owns them. Load only the guide or script required for the current
 - [`bottom-up-analyst`](bottom-up-analyst/) contains valuation arithmetic and memo frameworks.
 - [`pitch-like-lou`](pitch-like-lou/) contains the pitch-writing workflow and reference corpus.
 
-Each skill's README documents its own dependencies and usage. Runtime caches are generated
-next to the relevant workspace and are git-ignored.
-
-## Current snapshot
-
-The following commands measure the entry-point surface and the explicitly referenced skill
-resources from the repository root:
-
-```bash
-cloc --by-file --include-lang=Markdown \
-  skills/bottom-up-analyst/SKILL.md \
-  skills/pitch-like-lou/SKILL.md \
-  skills/sec-edgar-skill/SKILL.md \
-  skills/signal-sweep/SKILL.md \
-  skills/market-scout/SKILL.md
-```
-
-```bash
-cloc \
-  skills/bottom-up-analyst/SKILL.md \
-  skills/bottom-up-analyst/references/memo_template.md \
-  skills/bottom-up-analyst/references/guide_normalization.md \
-  skills/bottom-up-analyst/references/guide_competitive.md \
-  skills/bottom-up-analyst/references/guide_valuation.md \
-  skills/bottom-up-analyst/references/guide_ownership_signals.md \
-  skills/bottom-up-analyst/references/archetypes/*.md \
-  skills/bottom-up-analyst/scripts/dcf.py \
-  skills/bottom-up-analyst/scripts/epv.py \
-  skills/market-scout/SKILL.md \
-  skills/market-scout/scripts/fetch_market_data.py \
-  skills/market-scout/scripts/fetch_transcripts.py \
-  skills/pitch-like-lou/SKILL.md \
-  skills/pitch-like-lou/references/corpus/*.md \
-  skills/sec-edgar-skill/SKILL.md \
-  skills/sec-edgar-skill/references/guide_core.md \
-  skills/sec-edgar-skill/references/guide_filings.md \
-  skills/sec-edgar-skill/references/guide_financials.md \
-  skills/sec-edgar-skill/references/guide_ownership.md \
-  skills/sec-edgar-skill/references/guide_proxy.md \
-  skills/sec-edgar-skill/references/guide_holdings.md \
-  skills/sec-edgar-skill/scripts/orient.py \
-  skills/sec-edgar-skill/scripts/fetch_filing.py \
-  skills/sec-edgar-skill/scripts/fetch_filings.py \
-  skills/sec-edgar-skill/scripts/parse_financials.py \
-  skills/sec-edgar-skill/scripts/list_headings.py \
-  skills/sec-edgar-skill/scripts/fetch_insider_trades.py \
-  skills/sec-edgar-skill/scripts/fetch_13f_holders.py \
-  skills/sec-edgar-skill/scripts/test_setup.py \
-  skills/signal-sweep/SKILL.md \
-  skills/signal-sweep/screens.json \
-  skills/signal-sweep/references/guide_screens.md \
-  skills/signal-sweep/scripts/scan_insiders.py \
-  skills/signal-sweep/scripts/scan_market.py \
-  skills/signal-sweep/scripts/search_themes.py \
-  skills/signal-sweep/scripts/scan_conferences.py
-```
+Each skill's README documents its dependencies and usage. Invoke installed artifact-producing
+scripts from the research workspace so their git-ignored runtime caches stay beside the work.
 
 ## Research scope
 

skills/market-scout/README.md

@@ -1,41 +1,41 @@
 # Market Scout
 
-Pull public market data — price, market cap, trailing returns, peers, and sector screens — for
-US-listed stocks via [`yfinance`](https://github.com/ranaroussi/yfinance). An unopinionated data
-layer: it surfaces facts and rankings; it decides nothing.
+Public market context and earnings-call transcript retrieval for US-listed stocks via
+[`yfinance`](https://github.com/ranaroussi/yfinance) and Yahoo Finance. The skill returns source
+material; it does not decide whether a security is attractive.
 
 Part of the [SecStack skills](../README.md) collection.
 
-## Installation (do this first)
+## Setup
 
-Install this skill's dependencies and the browser runtime before using the scripts:
+The SecStack bootstrap installs this skill's dependencies into the profile environment. For
+standalone use:
 
 ```bash
-uv sync
-# The SecStack bootstrap installs agent-browser into the isolated Pi profile.
-# Run this once from an active SecStack profile:
-agent-browser install
+uv sync --project "<skill-dir>"
 ```
 
-- `yfinance`/`pandas` power market/peer data.
-- `agent-browser` is required for earnings-call transcript pages (JS-rendered Yahoo/Quartr).
-- In the packaged SecStack profile, its Pi-managed binary is already on PATH.
+Activate the resulting environment or prefix script commands with
+`uv run --project "<skill-dir>"`.
 
-## Use
+Transcript retrieval also uses the Pi-managed `agent-browser` binary. Install its browser runtime
+once from an active SecStack profile:
 
 ```bash
-python scripts/fetch_market_data.py --ticker AAPL --peers
+agent-browser install
 ```
 
-## Earnings call transcripts (Yahoo + Quartr)
+## Examples
+
+Keep the research workspace as the current directory and invoke the installed scripts by their
+resolved paths so generated transcript cache stays with the research:
 
 ```bash
-python scripts/fetch_transcripts.py --ticker AAPL --list
-python scripts/fetch_transcripts.py --ticker AAPL --latest 1
+python "<skill-dir>/scripts/fetch_market_data.py" --ticker AAPL --peers
+python "<skill-dir>/scripts/fetch_transcripts.py" --ticker AAPL --list
+python "<skill-dir>/scripts/fetch_transcripts.py" --ticker AAPL --latest 1
 ```
 
-yfinance offers far more than the script wraps and is self-documenting (`dir()`, `help()`,
-`t.info.keys()`) — see [SKILL.md](SKILL.md) for how to discover and construct what you need.
-
-Market data is best-effort and occasionally stale or missing for thinly-covered names; confirm
-anything load-bearing against a primary source.
+See [SKILL.md](SKILL.md) for routing, output contracts, and runtime discovery beyond the bundled
+report fields. Yahoo data is best-effort and can be stale or incomplete for thinly covered names;
+verify load-bearing facts against an issuer or regulatory source.

skills/market-scout/SKILL.md

@@ -1,81 +1,85 @@
 ---
 name: market-scout
 description: >-
-  Pull public market data for US-listed stocks via Yahoo Finance: price, market cap, shares,
-  52-week range, trailing returns, sector/industry peer tables and pre-ranked screens, and
-  earnings call transcripts. Use this whenever a task needs a quick market snapshot or quote
-  for a ticker, trailing performance, a company's peer set, a sector/theme shortlist, or
-  earnings call transcripts — e.g. "what's the price/return on X", "who are X's peers",
-  "best-performing names in this industry", "get me the latest earnings call", or turning a
-  theme into a concrete list of tickers. This is an unopinionated data layer; it does not
-  decide what is cheap, good, or worth buying.
+  Retrieve current public market data and earnings-call transcripts for US-listed stocks via
+  Yahoo Finance: quotes, market capitalization, shares, price history and trailing returns,
+  industry context, peer tables, and transcript text. Use when a task needs a current market
+  snapshot, performance calculation, Yahoo peer or industry data, a market-data field, or an
+  earnings-call transcript. Use primary filings for facts that Yahoo does not author or when a
+  load-bearing figure needs regulatory verification.
 ---
 
 # Market Scout
 
-Pull **public market data**, **sector/peer screening**, and **earnings call transcripts**
-for US-listed stocks. Unopinionated: it surfaces prices, returns, peers, rankings, and
-management commentary; it does not decide what is cheap or worth buying — leave that to
-whatever framework is driving.
+Retrieve market context and transcript text without turning the result into an investment
+conclusion.
 
-## Setup (install first)
+## Runtime and paths
 
-Install this skill's dependencies from this directory before running any script:
+Resolve bundled paths relative to this `SKILL.md`. Invoke scripts by absolute path while keeping
+the shell working directory at the research workspace. This puts the default
+`./transcript-cache` beside the work rather than inside the installed skill; alternatively pass
+`--cache-dir`.
 
-```bash
-uv sync
-# In the packaged SecStack profile, agent-browser is installed by the bootstrap.
-agent-browser install
-```
+The packaged SecStack profile already exposes this skill's Python dependencies. For standalone
+use, run `uv sync --project "<skill-dir>"`, then either activate that environment or prefix the
+examples below with `uv run --project "<skill-dir>"`. `fetch_transcripts.py` also requires the
+`agent-browser` binary and its one-time browser installation (`agent-browser install`). No API key
+or SEC identity is required.
 
-- `agent-browser` is required for earnings-call transcripts (`fetch_transcripts.py`).
-- No API key or identity is needed — Yahoo Finance is public.
+## Choose a route
 
-## Market snapshot and peers
+### Snapshot, returns, industry, and peers
 
-`fetch_market_data.py` prints a compact Markdown summary to stdout — price, market cap,
-shares outstanding, 52-week range, trailing returns, and (optionally) industry overview and
-peer tables. Output is live and never cached. `--help` is the authoritative flag reference:
+`fetch_market_data.py` prints a live Markdown report to stdout. It does not cache time-sensitive
+market data.
 
 ```bash
-python scripts/fetch_market_data.py --ticker AAPL --industry --peers
+python "<skill-dir>/scripts/fetch_market_data.py" --ticker AAPL
+python "<skill-dir>/scripts/fetch_market_data.py" --ticker AAPL --industry --peers
 ```
 
-## Earnings call transcripts
+The default report includes common trailing windows available inside the requested history period.
+Use `--period` to change how much history is fetched. For a different interval or field, use the
+runtime-discovery route below rather than treating the bundled report as Yahoo's full schema.
+
+### Earnings-call transcripts
 
-`fetch_transcripts.py` scrapes Yahoo Finance's Quartr-powered transcript pages via
-`agent-browser` (Yahoo requires JS rendering). It lists available transcripts or downloads
-them as LLM-friendly Markdown to `transcript-cache/<TICKER>/transcripts/`. Files are
-named `Q3-FY2026.md` etc., cached and reused across runs. `--help` for all flags:
+List the periods Yahoo currently exposes, then request the exact period or latest count needed:
 
 ```bash
-python scripts/fetch_transcripts.py --ticker AAPL --list           # list available
-python scripts/fetch_transcripts.py --ticker AAPL --latest 1       # most recent
-python scripts/fetch_transcripts.py --ticker AAPL --year 2025      # full fiscal year
-python scripts/fetch_transcripts.py --ticker AAPL --quarter Q3 --year 2025
+python "<skill-dir>/scripts/fetch_transcripts.py" --ticker AAPL --list
+python "<skill-dir>/scripts/fetch_transcripts.py" --ticker AAPL --latest 1
+python "<skill-dir>/scripts/fetch_transcripts.py" --ticker AAPL --year 2025
+python "<skill-dir>/scripts/fetch_transcripts.py" --ticker AAPL --quarter Q3 --year 2025
 ```
 
-Each cached file has a summary, `## Prepared Remarks` with `### Speaker — Title` headings,
-and a `## Q&A` section — greppable by speaker name, "guidance", "margin", or any keyword.
+Downloads are written to `<cache>/<TICKER>/transcripts/Q3-FY2026.md` and reused on later runs.
+Artifact-producing mode emits one absolute path per completed transcript to stdout and diagnostics
+to stderr. A partial or total retrieval failure exits nonzero rather than masquerading as an empty
+period.
 
-## Beyond the bundled scripts — yfinance is self-documenting
+Each file preserves the Yahoo source URL and, when present, separates prepared remarks from Q&A.
+Transcript text and speaker attribution are third-party data; verify a consequential quote against
+the issuer's own transcript, webcast, or filing when available.
 
-The scripts above wrap the common cases. yfinance exposes far more (financials, holders,
-options, earnings dates, calendar, sector/industry screens, …). When a task needs something
-the scripts don't cover, **discover at runtime** rather than guessing field names:
+## Runtime discovery beyond the wrappers
+
+The bundled scripts cover frequent jobs, not the limits of `yfinance`. Inspect the installed API
+instead of guessing field names or assuming a fixed metric template:
 
 ```python
 import yfinance as yf
 
-t = yf.Ticker("AAPL")
-
-print([a for a in dir(t) if not a.startswith("_")])  # all attributes/methods
-list(t.info.keys())  # every field in the snapshot
+ticker = yf.Ticker("AAPL")
+print([name for name in dir(ticker) if not name.startswith("_")])
+print(sorted((ticker.info or {}).keys()))
+help(ticker.history)
 
-# Sector/industry screening (theme -> shortlist):
-ind = yf.Industry(t.info["industryKey"])
-print([a for a in dir(ind) if not a.startswith("_")])  # top_companies, overview, ...
+industry = yf.Industry((ticker.info or {})["industryKey"])
+print([name for name in dir(industry) if not name.startswith("_")])
 ```
 
-When a field or method isn't what you expected, `dir()` / `help()` / `.info.keys()` recover
-the answer inline — prefer that over guessing.
+Use this route for calendars, options, holders, financial tables, custom return windows, or other
+Yahoo fields. Report the field name, period, units, and retrieval date. Treat missing or stale data
+as missing; do not silently substitute a different field or period.

skills/market-scout/scripts/_common.py

@@ -1,17 +1,30 @@
-"""Minimal shared runtime setup for market-scout scripts."""
-
-import sys
-
-# yfinance/pandas can emit non-ASCII text (e.g. company names like "Société");
-# force UTF-8 so a Windows cp1252 console doesn't raise UnicodeEncodeError.
-if sys.platform.startswith("win"):
-    for _stream in (sys.stdout, sys.stderr):
-        try:
-            _stream.reconfigure(encoding="utf-8")
-        except Exception:
-            pass
-
-
-def log(msg: str) -> None:
-    """Progress/diagnostics -> stderr (keeps stdout clean for the result)."""
-    print(msg, file=sys.stderr, flush=True)
+"""Shared runtime and output helpers for market-scout scripts."""
+
+import os
+import sys
+from pathlib import Path
+
+# Provider data can contain non-ASCII company and speaker names.
+if sys.platform.startswith("win"):
+    for _stream in (sys.stdout, sys.stderr):
+        try:
+            _stream.reconfigure(encoding="utf-8")
+        except Exception:
+            pass
+
+try:
+    import truststore
+
+    truststore.inject_into_ssl()
+except Exception:
+    pass
+
+
+def log(msg: str) -> None:
+    """Write human-readable progress to stderr."""
+    print(msg, file=sys.stderr, flush=True)
+
+
+def emit(path: str | os.PathLike) -> None:
+    """Write one absolute artifact path to stdout."""
+    print(str(Path(path).resolve()), flush=True)