Add agentict CLI: one-time PESTLE signal monitoring for market watchlists

[Copilot]

Investors monitoring a watchlist currently have to manually crawl news/financial sites for PESTLE signals before deciding whether to invest. This adds a Python CLI (agentict) that runs once, reads a Markdown watchlist, gathers signals per (ticker, exchange), and emits a single tabular report ending in an Invest / Not / no data available verdict.

Watchlist parsing

• New parser.py validates a # Stock exchange / # Tickers grammar; missing or misordered headings hard-fail the whole run (no partial report).
• In-section duplicate tickers collapse to one entry; a ticker declared under 2+ exchanges is preserved as one row per exchange, each forced to no data available with a "duplicate across exchanges" reason, and skips signal collection entirely.
• Exchange sections with zero tickers are valid and contribute no rows.

Signal collection

• sources/ defines a SignalSource protocol with two collectors (web-search style, Yahoo-Finance style) built on requests, each wrapping every failure (network, timeout, parse) into SourceError so one source's failure never aborts a ticker's processing.
• Orchestrator proceeds with best-effort partial signals if some sources fail; only forces no data available when all sources fail for a ticker.

Verdict via pluggable Financial Analyst agent

• agents/base.py defines a FinancialAnalyst interface (assess(ticker, exchange, signals) -> VerdictResult) so the verdict is judgment-based rather than a fixed scoring formula.
• Default HeuristicFinancialAnalyst is deterministic and network-free (documented as an illustrative stand-in, not a certified model); it discretionarily returns no data available when fewer than 2 of 6 PESTLE categories have usable signal or the signal is materially conflicting.
• An optional agents/llm.py extension point is selectable via --analyst llm / AGENTICT_ANALYST=llm, with lazy SDK import so it's never a hard dependency.
• Signal aggregation keeps uncategorized source text in a distinct bucket rather than broadcasting it across all 6 PESTLE categories, so a single uncategorized source can't trivially satisfy the multi-category coverage check or skew keyword tallies.

Report and CLI

• report.py renders a plain-text Ticker/Exchange/Verdict/Reason table and always prepends a mandatory AI-generated/informational-only disclaimer.
• agentict monitor --file <path>.md [--output <path>] [--analyst heuristic|llm]; exit codes: 0 success, 2 usage/malformed-watchlist error (no report emitted), 1 unexpected error.

Packaging and CI

• pyproject.toml (hatchling, src/ layout, requires-python>=3.10, only runtime dependency is requests), console script agentict.
• .github/workflows/python-ci.yml runs pytest on ubuntu-latest and windows-latest.

$ agentict monitor --file watchlist.md
DISCLAIMER: This report is generated by an automated AI system...

Ticker | Exchange | Verdict           | Reason
-------+----------+-------------------+------------------------------------------
AAPL   | NASDAQ   | Invest            | Positive keyword signal (3) outweighs...
GOOGL  | NASDAQ   | no data available | duplicate across exchanges: 'GOOGL'...

Includes unit test coverage for the parser grammar, dedup/cross-exchange handling, orchestrator failure modes, heuristic agent branches, and report rendering.

• Fixes [Product Request] Enable one-time PESTLE signal monitoring for market watchlists via CLI #4

[olileger]

@copilot update the prject's README file to explain how to use that new feature (installation, run example)

[Copilot]

@copilot update the prject's README file to explain how to use that new feature (installation, run example)

Added a "agentict CLI: one-time PESTLE signal monitoring" section to the README covering installation, watchlist format, usage, and exit codes.