Skills agent-sheet

shell-native spreadsheet CLI for agent work. Use it for reconnaissance, search, fill propagation, programmable workbook logic, and bulk data roundtrips with explicit verification.

install
source · Clone the upstream repo
git clone https://github.com/dream-num/skills
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/dream-num/skills "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/agent-sheet" ~/.claude/skills/dream-num-skills-agent-sheet && rm -rf "$T"
manifest: skills/agent-sheet/SKILL.md
source content

agent-sheet

agent-sheet
is a local-first spreadsheet CLI for agents. Route tasks through the canonical public surface:

  • inspect
    for reconnaissance
  • search
    for first-class localization
  • fill
    for first-class propagation
  • run
    for default programmable workbook work
  • pipe out
    /
    pipe in
    for bulk rectangular data transfer

Prefer it when the task needs real workbook structure, formulas, workbook-local logic, or a safe shell roundtrip instead of ad hoc CSV-only handling.

Use it for

  • resolving a workbook and keeping the target explicit with
    --entry-id
  • inspecting workbook structure, ranges, formulas, and lint signals
  • localizing hits before a mutation with
    search
  • propagating formulas or series with
    fill
  • importing a local workbook, continuing edits, and exporting a final file
  • streaming workbook data through shell tools with
    pipe out
    /
    pipe in
  • using bounded
    run
    logic when the smaller primitives do not express the task cleanly

Do not default to

  • reopening the workbook with another local workbook library for work
    agent-sheet
    already covers
  • using
    run
    when
    inspect
    ,
    search
    ,
    fill
    , or
    pipe
    already gives a smaller, clearer surface
  • trusting entry metadata alone as proof of workbook-visible structure
  • finishing after a mutation without a task-specific verification pass

First path

  1. Run playbooks/00-preflight.md.
  2. Resolve the workbook and keep using
    --entry-id
    .
  3. Choose the smallest matching lane via references/command-selection-matrix.md.
  4. Verify with playbooks/15-verify.md before finishing.

Hard defaults

  • prefer
    entryId
    over
    unitId
    for local workbooks and imports
  • treat
    inspect
    as the default first step for reconnaissance
  • treat
    search
    as the default hit-localization surface
  • treat
    fill
    as the default propagation primitive when the workbook already has a correct seed
  • treat
    run
    as the default programmable workbook surface
  • treat
    pipe out
    /
    pipe in
    as stdin/stdout-first bulk transfer
  • verify shell roundtrips with structure plus sample rows, not row count alone
  • use
    inspect workbook
    for workbook-visible structure and handoff verification

Highest-signal gotchas

  • imported local entries can be healthy even when later
    file info
    shows
    unitId: null
    ; keep operating on
    entryId
  • do not use
    agent-sheet
    as the tool for "user directly edits origin workbook"; that actor should be a real remote user in the browser
  • file info
    is metadata only; it does not prove worksheet count, worksheet names, or formula state
  • non-English worksheet names work, but quote the full A1 range string in the shell
  • shell pipelines can preserve the expected row count while still shifting headers or keys; verify the first rows and key columns after writeback
  • if the same
    run
    fallback repeats, that is product-surface feedback rather than a permanent prompt habit

Read references/gotchas.md when the task looks routine but has import/handoff, shell roundtrip, or workspace ambiguity.

Task routing

TaskRead next
workspace root, CLI resolution, or workbook target is unclearplaybooks/00-preflight.md
inspect workbook state, extract bounded data, search, or review formulasplaybooks/10-read-analyze.md
plan the smallest safe mutation pathreferences/command-selection-matrix.md
make data-visible edits or workbook-local changesplaybooks/20-write-safe.md
verify a mutation, shell roundtrip, or handoffplaybooks/15-verify.md
create, import, open, or export a local workbookplaybooks/30-file-lifecycle.md
stream workbook data through shell toolsreferences/shell-patterns.md
formatting, layout, or other primitive gapsplaybooks/40-script-fallback.md
import/handoff,
entryId
targeting, metadata-only checks, or shell verification edge cases
references/gotchas.md

Output style

  • prefer bounded previews, file paths, or stream output over oversized inline dumps
  • use
    pipe out
    for shell pipelines
  • use explicit artifacts when the result is large or reused
  • when reporting success, include the verification surface you actually used

Ready-made assets

Use these before improvising shell snippets:

AssetUse when
examples/roundtrip-pipe-review-rectangle.mdbuilding a review-style dataset through shell roundtrip
examples/handoff-verify.mdexporting, importing, and proving handoff structure
examples/template-import-anchor-check.mdimporting a template workbook and checking anchor cells
scripts/verify_csv_preview.pycomparing header, head sample rows, and a key column
scripts/check_csv_cells.pyasserting anchor values or non-empty cells from a CSV snippet

Read next

Read only the file needed for the task:

FileUse when
playbooks/00-preflight.mdworkspace or workbook context is not yet resolved
playbooks/10-read-analyze.mdinspecting workbook state, extracting data, searching, or reviewing formulas
references/command-selection-matrix.mdpicking the smallest correct command before a mutation
playbooks/15-verify.mdproving a mutation or handoff actually succeeded
playbooks/20-write-safe.mdchoosing the smallest safe mutation path
playbooks/30-file-lifecycle.mdcreating, importing, opening, or exporting a local workbook
playbooks/40-script-fallback.mdbuilt-in primitives cannot express the requested workbook change
references/shell-patterns.mdthe task is naturally a shell pipeline
references/gotchas.mdthe task involves common real-world failure modes
references/js-api-minimal.md
run
is necessary and must stay tightly bounded
examples/roundtrip-pipe-review-rectangle.mdyou need a concrete shell roundtrip example
examples/handoff-verify.mdyou need a concrete export/import handoff example
examples/template-import-anchor-check.mdyou need a concrete non-English template import example