automate-hancom-documents

automate-hancom-documents is a skill for Codex from innae1121-bit/gsghwp. It costs 98 tokens per session (7,039 once invoked), scanned A, original, MIT.

A routing skill for editing, inspecting, analysing, and verifying Hancom HWP documents. Hancom HWP is a document format and editor commonly used in Korea.

In plain words
What is it for?
It is for inspecting or changing HWP documents, including tables, captions, and images, as well as working with related PPTX and XLSX content.
Why use it?
It keeps HWP work on the required document-automation connection and requires confirmation that a requested change actually succeeded.

Skill for Codex

Written for Codex: agents/openai.yaml present. Also seen: mentions Codex.

Good fit It is for inspecting or changing HWP documents, including tables, captions, and images, as well as working with related PPTX and XLSX content.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/innae1121-bit/gsghwp/automate-hancom-documents
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent
npx skills add innae1121-bit/gsghwp --skill automate-hancom-documents
Clone the repo
git clone --depth 1 https://github.com/innae1121-bit/gsghwp

Made for: Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for automate-hancom-documents

README.md
[![agentmods](https://agentmods.dev/badge/skills/innae1121-bit/gsghwp/automate-hancom-documents/github.svg)](https://agentmods.dev/skills/innae1121-bit/gsghwp/automate-hancom-documents)
Your own site
<a href="https://agentmods.dev/skills/innae1121-bit/gsghwp/automate-hancom-documents"><img src="https://agentmods.dev/badge/skills/innae1121-bit/gsghwp/automate-hancom-documents/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for automate-hancom-documents

Your own site · 80×15
<a href="https://agentmods.dev/skills/innae1121-bit/gsghwp/automate-hancom-documents"><img src="https://agentmods.dev/badge/skills/innae1121-bit/gsghwp/automate-hancom-documents.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 98 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 7,039 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector warn 7 Sept 2026
SkillSpector: 3 findings, up to high

These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →

  • high Anti-Refusal · line 157
    Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.
    Fix: Remove instructions that suppress warnings, disclaimers, or ethical commentary. Let the agent surface safety-relevant caveats to the user.
  • high Memory Poisoning · line 169
    Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.
    Fix: Protect agent memory and state from modification by untrusted content. Use read-only memory for critical instructions and validate all state changes.
  • medium Output Handling · line 93
    Output size or generation rate is not bounded. Unbounded output enables denial-of-service through resource exhaustion, log flooding, or context-window stuffing.
    Fix: Set explicit limits on output length, generation count, and rate. Use max_tokens and truncation to prevent unbounded output.
How audits are shown
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00098 $0.07039
Opus 5 $0.00049 $0.03520
Sonnet 5 $0.00020 $0.01408
Haiku 4.5 $0.00010 $0.00704

Measured 11d ago against content hash 240c63ee012b, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-10, from the pricing page.

Security

Grade A, and why

automate-hancom-documents scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.

The scan reads SKILL.md. This mod also ships 39 executable files (scripts/_hwp_hgn1_semantic_parity_codec.py, scripts/_hwp_hgn1_semantic_parity_models.py, scripts/_hwp_native_graph_errors.py, …), listed below but not scanned — reading those needs a real analyzer, not pattern matching.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

plugins/gsg-hwp/skills/automate-hancom-documents/SKILL.md · 170 lines

How it starts

The opening of the file, as written. The whole thing — 170 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Hancom Document Automation

Start with the production action path: match the request to one public tool and call it. Read a reference only when the work in front of you needs what that reference holds.

Images in a request

Read a visible value directly and call its matching write. For image or PPT structure reconstruction, read references/native-layout.md (identical MCP resource: gsg-hwp-beta://reference/native-layout). That reference is the sole reconstruction workflow; this entrypoint only routes to it.

Non-negotiable rules

  • Route every Hancom/HWP edit, analysis, and verification request through the gsg-hwp-beta-live MCP and this skill.
  • Treat a change as successful only after the selected public tool reaches a terminal success result and its operation-specific readback confirms the requested state. A dispatched call or mutation attempt without that readback is not success.
  • Choose the most specific public action tool for the user's request. Never invent an API name, cell address, control ID, or hidden execution field.
  • Before every public write, choose a caller-owned operation_id (for example, a UUID) and keep it unchanged until that logical operation reaches a terminal result. Reuse the same ID only when retrying the same payload or recovering a lost response; use a new ID for an intentional second execution, even when its payload is identical. Never derive the ID from the payload. The same ID with different inputs is a conflict. Completed results remain replayable for up to 30 days by default; when the terminal journal exceeds 10,000 records or 64 MiB, the oldest terminal results are evicted first. Never recycle an ID after journal expiry or a not_found response.
  • Live mode attaches only to the exact active, saved, editable HWP already registered in ROT. It does not open, save, clear, quit, or change history by default. Use hwp_undo or hwp_redo only for an explicit history request, and hwp_save_reopen_verify only for an explicit save-and-reopen verification request.
  • Never use computer-use, Windows UI automation, keyboard/mouse control, or a Python COM document call as a fallback for an HWP operation. If the gsg-hwp-beta-live MCP is unavailable, disconnected, or returns a closed transport, report that exact state and stop. If a certified native operation is unavailable, return the structured result from the selected HWP tool and stop.
  • Do not hard-code a document title, page number, table size, or cell map in reusable code.
  • When the user asks to analyze or verify how an open HWP page looks, call hwp_render_page; it preserves document state and uses HWP CreatePageImage. Do not render in the middle of a fill, repeat, sync, or Excel append. For cell text, IDs, merges, or sizes use hwp_inspect_page_fast with include_cells=true. hwp_inspect_structure is not the cell-text readback.
  • hwp_render_page publishes each completed PNG at a unique path only after CreatePageImage finishes. Active session directories are protected by a worker ownership lock. Released results have a 24-hour TTL and at least a 15-minute client-read grace; cleanup runs at worker startup and every 15 minutes, removes abandoned partial files, and evicts the oldest eligible results when the preview store exceeds 64 MiB.
  • Document-local styles are read with hwp_list_styles; never use web search to infer a style ID or local title convention.
  • For an explicit convention-matching write, read observed_usage.convention_profile once and reuse only values the target document actually reports. Check each axis's sample_count, consistency, complete, variants, scope, and write_fields; never turn a conflicting or partial sample into a universal template.
  • An unobserved convention axis is missing evidence, not a reason to refuse the edit. Leave that write field unchanged unless the user supplied it explicitly, and continue with the observed axes the write tool supports.
  • Table-series tools copy the first table, regenerate each copied automatic-number control, and update only the literal title suffix while preserving the source caption style. Do not follow a successful series call with manual hwp_add_caption calls. If caption verification fails, report that failure instead of retrying captions or Undo in a loop.
  • After a plugin update, stay in the current Codex task and call hwp_runtime_info. The stable MCP proxy reloads the worker automatically when its source hash changes and emits a tool-list change notification; call hwp_reload only to force a reload. Compare mcp, worker_process_id, source_path, and tool_schema_hash before continuing. native_bridge is only the version this build expects; loaded_native_bridge is the bridge DLL the running Hangul editor actually has mapped, read from its module list, and it is only ever asked of Hwp.exe — Hancom's HwpApi.exe automation host never loads the bridge and is not evidence about it. The bridge verdict is reported, never enforced: it does not gate any operation and never by itself makes you stop, refuse, or restart Hangul. Continue the requested work and pass native_bridge_notice through to the user when native_bridge_state is anything other than matched. Only native_bridge_state="mismatched" (hangul_restart_required=true) proves a stale DLL is executing right now; not_loaded means that Hangul started before this bridge was registered and the next Hangul start picks it up on its own. Never restart Hangul on your own initiative — the user may have unsaved work open; state the notice and let the user decide. If the refreshed direct tool is not yet visible, call it through hwp_execute with the exact tool name and arguments.
  • The host may expose this skill directly from the plugin manifest, from a host cache, or through a legacy global Junction. Read the exact displayed SKILL.md path directly without assuming its storage form; do not locate, search, or compare other cache or installation paths.
  • For table data keyed by visible header labels, use records. Use cells only with verified cell addresses and rows only when the user supplied an explicit start_cell. A returned target_id changes only the target, never the chosen data format.
  • Explicit 1 mm table rows and columns are supported. To make the physical cell reach that size, set every affected cell to font_size_pt=1, line_spacing_percent=50, and zero padding; larger inherited text or padding can make HWP expand the cell. HWP stores 1 mm as 283 HWPUNIT, so inspection reports about 0.998 mm per row or column.
  • For an irregular or partially merged table, use the inspected CellTopology: actual cell address, row_span, column_span, and physical right/down neighbours. Never synthesize F1, A11, or an address-distance move sequence. Column and row formatting starts at the requested real cell; merging proceeds only through the recorded neighbour path after the covered owners form a gap-free rectangle.
  • hwp_split_table_cell has two deliberately different modes. Use split_mode="equal" only to subdivide one unmerged physical cell. Use split_mode="existing_grid" only to restore the exact row_span × column_span grid hidden by a merged cell. Do not split a merged header merely to reach a column for formatting; format the real intersecting cell instead. A requested existing-grid split that does not match the inspected spans must be rejected before mutation.
  • hwp_apply_page_plan keeps each block's absolute box.top_mm. It does that by writing the planned gap into the paragraph spacing of the block below it, which is the only vertical control a flow layout has; that channel is filled by the compiler and is deliberately absent from the published layout schema, so a layout tool cannot send it and a plan cannot ask for placement by hand. Blocks are emitted top-first regardless of declaration order. A block whose plan box overlaps the one above it, or whose content is taller than its box, is pushed down rather than moved up, and everything after it is measured from where it really ends. Positions land within about 0.2 mm; a plan whose content genuinely exceeds the body is rejected with the millimetres it is short.
  • Use hwp_insert_layout for live insertion away from the document end. layout.target=current inserts at the user's collapsed cursor. layout.target=after_page with page=N inserts at the start of the existing next page and adds one trailing page break, so it creates exactly one isolated page without a blank page; after the last page it appends with one leading page break. Do not add another page_break block. Continue to use hwp_append_layout only for the document end.
  • For screenshot or reference-layout reconstruction, restore the evidence-first sequence: call hwp_analyze_reference_image once with full_scan=true, inspect its overlay/contact sheets, let the model choose semantic elements and loose raster boxes, then call hwp_prepare_image_crops once. Mark rectangular photos as tight_raster and annotated maps/drawings/logos as annotated_group; crop preparation resolves the final boundaries and returns a boundary overlay while the model still owns meaning. In native tables, apply visible-surface ownership recursively: group title bands, repeated-item subheading pills, and their body boxes are separate cells or rows, never line-broken text inside one cell.
  • Never delete and rebuild a page for a local correction. Use hwp_patch_text for a sentence or line break, table formatting tools for an existing ordinary table, image replace/size paths for one picture, and reference_layout_patch for existing reference-grid rows, columns, styles, or edges. Rebuild only when structural inspection proves that no local target can represent the requested correction.
  • headers filters target tables; it is not a copy of the records keys. Use only labels known to be visibly present in the target table. If a page already narrows the request sufficiently, pass page without speculative headers.

Read the full file on GitHub · 170 lines

Files

What ships with it

60 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 11d ago First seen · 170 lines · 98 tokens per session scan A 240c63ee012b

Subscribe to this mod's changes

automate-hancom-documents is a skill published in the GitHub repository innae1121-bit/gsghwp (101 stars, last pushed 10d ago), licensed MIT. It adds 98 tokens to every session and 7,039 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other skills, from other repositories

baoyu-youtube-transcript

A tool for downloading the written captions, subtitles, chapter information, speaker labels, and cover image from a YouTube video using its URL or ID.

JimLiu/baoyu-skills · 107 tokens

orbit-notion

Open Orbit briefing skill — selected by the Orbit pipeline when Notion is the user's only connected connector, or when the user explicitly scopes their daily digest to Notion. Pulls the past 24 hours of document edits, comments, mentions, and database row changes from the user's authenticated Notion connection and…

nexu-io/open-design · 117 tokens

instrument-data-to-allotrope

Convert laboratory instrument output files (PDF, CSV, Excel, TXT) to Allotrope Simple Model (ASM) JSON format or flattened 2D CSV. Use this skill when scientists need to standardize instrument data for LIMS systems, data lakes, or downstream analysis. Supports auto-detection of instrument types. Outputs include full…

anthropics/knowledge-work-plugins · 123 tokens

feishu

Work with Feishu or Lark bots, docs, sheets, bitables, approval flows, and OpenAPI/MCP setup without hardcoding credentials.

Hmbown/CodeWhale · 33 tokens

read

Reads URLs and PDFs by fetching source content, defaulting to concise summaries for plain read requests and clean Markdown when asked to convert, save, quote, cite, or feed downstream work. Use when users ask in any language to read, fetch, check, summarize, quote, cite, convert, or save a URL or PDF. Not for local…

tw93/Waza · 78 tokens

overleaf-sync

A two-way connection between a local paper folder and Overleaf, a web-based LaTeX editor for writing research papers. It lets you move changes between the local files and the shared Overleaf project.

wanshuiyin/Auto-claude-code-research-in-sleep · 97 tokens