More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Peer Comps Done Right: bands, outliers, honest peers

    1

    Pick the peer set, quote the right band, drop outliers properly, and see why the average lies. Includes an offline CSV check script.

    Free

    2 usesSecurity scanned
    Peer Comps Done Right: bands, outliers, honest peers

    Peer Comps Done Right: bands, outliers, honest peers

    Example session with this skill installed

    Compare SUBJ to its software peers in tests/sample.csv and tell me which cells are outliers.

    • Read your context and instructions
    • Compiled the peer comps done
    • Generated the data export

    Real output of python3 scripts/comps.py tests/sample.csv SUBJ on the bundled invented sample, 2026-11-16:

    Subject SUBJ | industry: Software | sector: Tech
    Peer set: industry, 6 names: AAA, BBB, CCC, DDD, EEE, FFF

    pe_ttm [x]: P25 25 | median 27 | P75 28 (n=5)
    loss-making, left out of band: FFF (rank last)
    SUBJ: 30 above P75, +11% vs median; rank 5 of 6 low-to-high
    OUTLIER EEE: 95 vs others' median 27, z +5.0, scale set by 0.5*|median| (13.5)
    ev_ebitda [x]: P25 17 | median 18 | P75 19 (n=5)
    SUBJ: 20 above P75, +11% vs median; rank 5 of 6 low-to-high
    op_margin [pct]: P25 0.242 | median 0.26 | P75 0.278 (n=6)
    SUBJ: 0.3 above P75, +15% vs median; rank 6 of 7 low-to-high
    roic [pct]: P25 0.2 | median 0.21 | P75 0.22 (n=5)
    SUBJ: 0.25 above P75, +19% vs median; rank 5 of 6 low-to-high

    Outlier means far from the rest, not wrong and not bad. Check the cause before dropping it.

    sample.csv

    CSV · data export

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Most peer comparisons go wrong before any arithmetic: the subject sits in its own benchmark, GOOG and GOOGL count as two companies, a loss-making name prints the "cheapest" EV/EBITDA, and one 95x P/E drags the average. This skill gives your agent the rules and the check.

    What it does

    Your agent builds the peer set (same industry when it has 5 or more names, else the sector), leaves the subject and duplicate share classes out, skips multiples at or below zero, quotes a P25 / median / P75 band with the count, refuses a band under 3 peers, and flags outliers with a leave-one-out robust z and an explicit floor. It also picks the multiple that fits the business (P/TBV for banks, FFO for REITs, normalised P/E for cyclicals).

    What you get

    PathHolds
    SKILL.mdten rules, a 15-row decision table, the method
    references/peer-selection.mdindustry vs sector, share classes, where candidate lists fail
    references/robust-stats.mdmedian + MAD, floors, worked example with invented numbers
    references/multiples-by-sector.mdwhich multiple for which business, and the traps
    templates/comps-note.mdcopy-paste write-up
    scripts/comps.pyoffline: CSV to bands, outliers, subject position
    tests/sample CSV and tests: happy path plus four refusals

    Who it's for

    Investors and analysts who want "cheap versus peers" to name the peers, the band and the sample size, and developers building a comps feature who need defensible defaults.

    Scope

    Covered: peer selection, relative-valuation bands, outlier handling, sector multiple choice, a CSV-driven check script. The thresholds (3 peers for a band, 5 for an industry, 3 robust sigma, 1.0x and 5 pp floors) are working defaults, not standards.

    Not covered: fetching data (bring your own CSV), DCF, private-company comps, precedent transactions.

    This is research tooling, not investment advice.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse All Claude Skills.

    No data fetching. Duplicate share classes are detected by identical figures on 4+ metrics, not by name. Thresholds are working defaults from one comps tool. The robust-z floors make the test strict on the low side of multiples.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    2 uses

    Times used to date

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 day ago

    • Free with an Agensi account

    Software
    Python 3.9+A CSV of peers you export yourself

    Permissions

    Runs one local python3 script on a CSV the user provides. No network.

    Listed1 day ago

    What's inside

    Frequently Asked Questions

    No. The script reads a CSV you supply (ticker, industry, sector and metric columns), so it runs offline with no keys.

    One extreme multiple moves the mean far more than the median. references/robust-stats.md shows it with invented numbers.

    The scale never drops under half the median, so for a positive multiple the flag effectively fires on the high side. The reference explains it and what to say when reporting.

    Yes: --sigmas, --scope and --unit are flags; the other defaults are constants at the top of scripts/comps.py.

    Free