Parallax Desk Call List

Desk Call List

When not to use

• Single portfolio morning analysis -> use /parallax-morning-brief
• One client meeting prep -> use /parallax-client-review
• Unweighted ticker surveillance -> use /parallax-watchlist-monitor
• Full stock drill-down, technicals, or analyst consensus -> use /parallax-should-i-buy or /parallax-watchlist-monitor
• Book creation/editing -> operator maintains ~/.parallax/desk-book/book.yaml

Gotchas

• JIT-load _parallax/parallax-conventions.md for §0.0 pre-flight, §0.1 tool loading, §0.2 typed integer params, §3 parallel execution, §3.1 annotation/rank separation, §9.1/§9.2 disclosures, §10 render gate, §11 verdict sensitivity, §12 information framing, and §13 audience mode.
• JIT-load _parallax/coverage-matrix.md before Batch A. export_price_series is equity-only and FREE, and its success:false response classifies ETFs; ETFs then price via etf_daily_price. This skill does not call etf_profile.
• JIT-load _parallax/house-view/loader.md §1-§2, §5, and §6. The house view annotates movers but never changes rank order or membership.
• JIT-load _parallax/white-label/integration-pattern.md before Pre-Render. Call load_rm_branding_context("desk call list", audience=) from _parallax/white-label/rm_consumer.py, passing the parsed audience= invocation flag (or None if absent). This visual-only seam blocks voice access and redacts source references.
• JIT-load references/desk-book-format.md before loading/validating any desk book.
• JIT-load references/ranking-and-bounding.md before thresholding, ranking, or rendering bounded sections.
• JIT-load references/talk-tracks.md before drafting per-client talk tracks.
• desk_call_list_logic.py is the pure arithmetic layer. Do not put MCP calls or local file writes in it.
• Client names and weights never go to Parallax MCP tools; only the deduplicated symbol union leaves the machine.
• SCAN_CONCURRENCY = 24 applies only to the wide price scan. If the live probe shows rate limiting, use 8.
• export_price_series is FREE; etf_daily_price costs 1 token per ETF priced. This skill no longer calls etf_profile. Do not invent numbers.
• client_safe is supported, but this is primarily an internal RM artifact.

Build one ranked, bounded morning call list for a relationship manager covering many client books.

Usage

/parallax-desk-call-list
/parallax-desk-call-list threshold=4.5 min_impact=0.4 detail_cap=6 subset="Northgate,NG-001"
/parallax-desk-call-list [{"client_name":"Northgate","portfolio":[{"symbol":"AAPL.O","weight":0.12,"asset_class":"equity"}]}] redact_names=true

Defaults:

• threshold=5.0 unless overridden by book default_threshold_pct
• min_impact=0.50 pp unless overridden by book default_min_impact_pp
• detail_cap=8, news_cap=8, audience=internal_analyst

Workflow

Execute using Parallax MCP tools plus local Python helpers.

Batch 0 - Pre-flight, Book, House View

1. Resolve canonical _parallax/... paths per _parallax/parallax-conventions.md §0.0.
2. Call ToolSearch with query "+Parallax" to load deferred Parallax MCP schemas before the first data call.
3. Load references/desk-book-format.md; read inline input or $PARALLAX_DESK_BOOK_PATH, else ~/.parallax/desk-book/book.yaml. Validate via desk_call_list_logic.py. Inline input wholly replaces the saved book. Preserve validation warnings for both input paths.
4. Apply subset filtering only to the saved book, passing the redact_names mode into resolution. Keep only global validation warnings and warnings owned by selected clients. Report unmatched names verbatim unless redact_names=true; in redacted mode retain and report only the unmatched count, then apply redact_names() so client refs and names are removed. If a non-empty subset matches no clients, refuse to scan under the same disclosure rule.
5. Compute staleness tier and the sorted deduplicated symbol union U.
6. Load active house view per _parallax/house-view/loader.md §1-§2. Shell out to view_status; do not recompute expiry date math. If absent or invalid, run without view.

If the first Parallax batch after tool loading is empty or cancelled, re-fire the whole batch once before concluding unavailability.

Batch A - Asset-Class Classification and Wide Price Scan

Load _parallax/coverage-matrix.md. Classification is fused with the price scan — the FREE export_price_series doubles as the equity classifier, so no separate classification probe is needed:

1. Cached asset_class → route directly: equity → export_price_series; etf → etf_daily_price. No probe.
2. Uncached → call export_price_series() (FREE, equity-only) as both classifier and equity price source:

• {"success": true, ...} with a non-empty prices list → equity; take the move and both close dates from the returned series. Cost 0.
• {"success": false, ...} / no prices → ETF; call etf_daily_price() for the move + per-row date.

Keyed on success == false (NOT an empty list — the tool returns a {success:false, error} object for ETFs). This is a deliberate, documented classifier, not the silent substitution coverage-lint guards against: a success:false that is actually a dead equity gets one etf_daily_price probe that also fails and lands in the unpriced/coverage handling, named in the coverage line.

1. Only if export_price_series is entirely unavailable (outage): resolve bare tickers to RICs per _parallax/parallax-conventions.md §1, then the static fallback (suffix-less or .P ⇒ ETF; otherwise equity).

Then fan out the price scan at SCAN_CONCURRENCY waves, plus one telemetry call:

| Tool | Parameters | Notes | |---|---|---| | get_telemetry | fields: regime_tag, signals, commentary.headline, divergences | Desk-wide market context | | export_price_series equity branch + classifier | symbol=, days=10 as typed integer | FREE. Equity-only; single-symbol calls; the success:false response also classifies ETFs | | etf_daily_price per ETF branch | symbol=, start_date=, end_date= | ETF-only; single-symbol calls; per-row date carries the move's session |

Derive move_pct = (close[-1] / close[-2] - 1) * 100 from the last two closes, which must both be finite. Record both dates — for the ETF branch these come from etf_daily_price rows (which carry the session date, unlike a dateless profile probe). If fewer than two closes return or either close/move is non-finite, mark the symbol unpriced, name it in coverage, and do not treat it as a trigger.

Batch B - Threshold and Scan Integrity

Load references/ranking-and-bounding.md.

1. Compute desk scan integrity before the empty-mover path. If priced coverage is below 80%, render SCAN DEGRADED and skip the call list.
2. Define movers as abs(move_pct) > threshold; the boundary is strictly greater.
3. If more than 40 symbols cross the threshold, auto-raise to the 40th-largest absolute move rounded up to the nearest 0.5 pp, recompute movers, and state the auto-raise.
4. If no movers, or movers exist but no client clears min_impact, render the no-calls short form.

Batch C - Mover Enrichment

For symbols in the mover set only, schedule enrichment in waves of at most 8 concurrent calls across symbols and tools:

| Tool | Parameters | Purpose | |---|---|---| | get_company_info | symbol for every mover | Display name and expected-name oracle | | get_peer_snapshot | symbol for every mover | Current factor scores and target_company | | get_score_analysis | symbol, weeks=4 as typed integer for equity movers only | Four-week score trajectory | | get_news_synthesis | symbol for top news_cap equity movers by desk-wide weighted exposure | One news synthesis per symbol, not per client |

Cross-check get_peer_snapshot.target_company against get_company_info.name. On mismatch, render Ground-truth Integrity, exclude that symbol's scores, but keep the price move in client ranking.

For ETF movers, skip get_score_analysis and get_news_synthesis because ETF coverage is unverified. Render those fields as not available for ETFs; never present their absence as an equity-grade enrichment failure. ETF price moves still drive client ranking.

Batch D - Client Ranking

Zero tool calls. Use desk_call_list_logic.py to compute triggered exposure, signed net impact, direction-agnostic severity, top driver, coverage, rank order, and bounded detail/summary sections. News and house-view tags are annotations only and must not alter rank membership or order.

Pre-Render - Load White-Label Branding

Load _parallax/white-label/integration-pattern.md §2. Import load_rm_branding_context from _parallax/white-label/rm_consumer.py. Call load_rm_branding_context("desk call list", audience=) once, passing the parsed audience= invocation flag (or None if absent) so the seam resolves §13.1 precedence over both the flag and the branding config. Place its header_lines below any House View Preamble. Place its about_lines in About This Report. Read resolved_audience off the returned context for the Output Format audience branches below (e.g. item 12's client_safe omission); do not re-resolve the mode yourself and do not inspect the branding mapping directly. The helper preserves the call list and selects default Parallax if branding is corrupt. The branding is for the desk's firm: one report, one brand.

Render - deterministic gate

Compose the complete report per Output Format, then run it through the shared render gate in one Bash step before replying:

DRAFT="$(mktemp "${TMPDIR:-/tmp}/deskcall.XXXXXX")"
cat > "$DRAFT" 
REPORT
python3 "/../_parallax/render_gate.py" --skill desk-call-list ** desk call list`.
3. `# Desk Call List` title with as-of date, session dates, threshold, clients scanned, and clients triggered.
4. Book Staleness Warning, only for warn/stale/unknown tiers.
5. Validation Warnings, whenever validation produced warnings, including renormalisation warnings for inline and saved books.
6. Ground-truth Integrity, only if any mismatch.
7. Market Context: 2-3 sentences from telemetry.
8. Priority Calls: render the informational preface per `_parallax/parallax-conventions.md §12`, then the full ranked table.
9. Client Detail: top `detail_cap` clients. Load `references/talk-tracks.md`; include why-listed arithmetic, positions table, talk track, and likely questions.
10. Also Affected (summary only): ranks after `detail_cap`; group by driving symbol when triggered clients exceed 25.
11. Symbol Movers Reference: symbol, name, move %, four-week score change, news headline, number of clients holding, desk-wide weighted exposure, and house-view tag.
12. Verdict Sensitivity per `_parallax/parallax-conventions.md §11`; omit when the seam's `resolved_audience` is `client_safe`.
13. Next steps: point to `/parallax-client-review`, `/parallax-should-i-buy`, or `/parallax-watchlist-monitor` as appropriate. Do not auto-invoke them.
14. About This Report: first the seam's `about_lines` verbatim — they already carry the `_parallax/white-label/integration-pattern.md §7` branding line, the unconditional currency line, any skipped local-logo basename, and the audience mode/notice lines; do not re-add any of them — then desk-book provenance, redaction state, and a cost note that the equity price scan is FREE (`export_price_series`) while `etf_daily_price` costs 1 token per ETF priced.
15. AI-interaction disclosure per `_parallax/parallax-conventions.md §9.2`.
16. Disclaimer: view-aware per `_parallax/house-view/loader.md §5` if active; otherwise render the standard disclaimer from `_parallax/parallax-conventions.md §9.1`.

## No-Calls and Degraded Short Forms

SCAN DEGRADED starts with `**Scan degraded — results not reliable.**` and names priced count, total symbols, coverage, and up to 10 unpriced symbols.

An empty client or symbol selection starts with `**Scan refused — no clients or symbols selected.**` and never renders a call list or no-calls result. An all-unmatched subset starts with `**Scan refused — subset matched no clients: .**`; under `redact_names=true`, replace the names with ` selector(s) matched no client`. Redacted refusal and partial-match output never contains raw selectors, and client refs are omitted everywhere.

Quiet mornings start with `**No calls indicated.**`, then unique symbols scanned, client books scanned, largest move, threshold, and the `min_impact` pp floor. Include House View Preamble, Branding Header, staleness warning, validation warnings, About This Report, AI disclosure, and disclaimer.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [bencharoenwong](https://github.com/bencharoenwong)
- **Source:** [bencharoenwong/parallax-workflows](https://github.com/bencharoenwong/parallax-workflows)
- **License:** MIT

Install and usage instructions live in the source repository linked above.