Borrowing it
Nothing to install: this file belongs to jakubsuplicki/codument. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/jakubsuplicki/codument/main/.agents/skills/work-step/SKILL.mdgit clone --depth 1 https://github.com/jakubsuplicki/codumentWrote 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.
[](https://agentmods.dev/skills/jakubsuplicki/codument/work-step)<a href="https://agentmods.dev/skills/jakubsuplicki/codument/work-step"><img src="https://agentmods.dev/badge/skills/jakubsuplicki/codument/work-step/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.
<a href="https://agentmods.dev/skills/jakubsuplicki/codument/work-step"><img src="https://agentmods.dev/badge/skills/jakubsuplicki/codument/work-step.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00021 | $0.03019 |
| Opus 5.5 | $0.00008 | $0.01208 |
| Sonnet 5.5 | $0.00004 | $0.00604 |
| Haiku 4.5 | $0.00002 | $0.00302 |
Grade A, and why
work-step 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 23d ago.
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.
How it starts
The opening of the file, as written. The whole thing — 116 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Work Step
Use this when the user says to continue, work the next step, or implement the approved plan.
Compatibility and project checks
At the first delivery action in a session, inspect codument --help from the executable you will
actually use: it must list work and verify. A version string alone does not establish capability.
If either is absent, repair the installation before executing this workflow: rebuild a linked
Codument checkout with npm run build, or install the matching package and run codument update.
Do not substitute older commands or invent manual approval/review checkpoints. Reuse this check
for the session unless the executable changes.
Run the project's existing relevant tests, lint and type checks. Codument's named-test reproduction
is a separate review check, not a requirement to install its own test runner. When reproduction is
needed, declare the project's file-targeted runner once with testCommand in .codument-meta.json
(including {file}); use a TAP reporter so failures are distinguishable from toolchain errors.
Workflow
- Read
codument work status --jsonand the selected plan underdocs/features,docs/concepts, ordocs/plans. Reconcile saved work with Git; hand its pending review or commit to the owning skill before starting another step. Explicitly resume interrupted work only when its resume condition is satisfied and the user has authorized continuation. - Confirm current bound approval with
codument steps --plan <path> --json. Record approval withwork approveonly after actual human approval. Foroutcome-v1, compare work with the approved outcome, milestone, constraints, acceptance evidence and verification; implementation routing discoveries alone do not require an amendment. Legacy file-bound plans retain their scope gate, and existing approvals cannot adopt outcome permission without renewed human approval. Approval does not mean paused work is running. - Start new selected work with
codument work start --plan <path>. Switching plans requires--reasonand preserves the old plan; usework supersedeonly for an explicit replacement decision. Pick the first unchecked step only after any saved delivery has passed its pending gate. - Surface the checklist in the live view (see Plan Checklist Mirror below): mirror the plan's steps into your host's native to-do panel with the step you are about to implement marked in progress, and run
codument steps --plan <active-plan> --emitsocodument watchshows the active step. Post that checklist inline in the chat as well — the step just completed, the step now starting, and what remains — because the native to-do panel and the watch tape are not the chat transcript, and a run that does not wait between steps otherwise advances with no in-chat marker. - Reuse the grounded planning context and read its owning docs, invariants and tests. If it is missing or stale, query the affected paths together with
codument context --paths <paths...>, or use--plan <active-plan>/--feature <slug>for the selected working set. Refresh after routing discoveries as needed; no ownership query is owed before and after every file. Read the whole registry when editing the map or when the CLI is unavailable. A context budget is soft: selected contracts may exceed it and must not be silently discarded. - Implement only that step toward its observable milestone. An infrastructure slice names the milestone it unblocks; do not claim that milestone complete without its promised evidence.
- Use
tddor the strongest practical verification loop. Diagnose routine test, mapping and documentation failures, correct their cause within the approved contract, and rerun the affected checks. A failed check prevents delivery while red; it does not by itself require human intervention. Use the repair boundary below. - Batch new source registration with
codument map materialize <files...>(see Feature Map Materialization), then update the mapped docs + registry as part of the same step. - Mark the step complete — in the plan doc, and in the mirrored native to-do list — only after implementation verification passes. Before final-step compaction, save the approved plan with a pending-review checkpoint at
.codument/pending-plans/<repo-relative-plan-path>; keep this recovery copy outside the staged step. Then compact the## Delivery Planblock perplan-with-docs(Compaction on ship): lift surviving decisions into## Decisions/ADRs and any newly-true constraint into## Invariants & boundaries, then delete the delivery scaffolding so the durable doc is left in the standard's layers. - Stage only the files belonging to this step. For the compacted final step, run
codument work finish --prepare-final, then stagedocs/.approvals.json; this binds the retained approved contract to the final staged delivery. Repeat preparation after any boundary correction. The ignored recovery copy is context, never the only approval evidence. Record the pending gate withcodument work resume --gate reviewbefore handing off. - Proceed directly to
review-workfor this step without waiting. It selects focused self-review or independent adversarial review from the generated minimum effort and the change's semantics. Never start the next delivery-plan step from here — review and commit come first, in either mode. - In gated mode, stop instead and present the user with end-of-step options:
- Run
review-worknow (recommended) - Make a specific correction to this step
- Pause here
- Run
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.
- 23d ago Changed · +15 lines f84bc424a6d5
- 1mo ago Changed · +19 lines bd36b2010aa3
- 1mo ago Changed · +9 lines 3274cedb7d97
- 1mo ago First seen · 73 lines · 21 tokens per session scan A 960a02ae29a9
work-step is a skill published in the GitHub repository jakubsuplicki/codument (48 stars, last pushed 23d ago), licensed Apache-2.0. It adds 21 tokens to every session and 3,019 once invoked, about $0.0001 per session on Opus 5.5. 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.
Other skills, from other repositories
do
Work a SpecWeave increment task by task - claim, implement inside its Files, test, commit, done with evidence - alone, unattended or in lanes, then close. Use for "implement", "next task".
handoff
Hand work to another tool, account or machine and pick it up there, in two words. Use when the user says "hand off", "hand off all", "out of tokens", "switching accounts" or "pick up".
sw-do
Work a SpecWeave increment task by task - claim, implement inside its Files, test, commit, done with evidence - alone, unattended or in lanes, then close. Use for "implement", "next task".
increment
Plan work as a SpecWeave increment - one spec.md with Problem, Scope, numbered ACs, Approach and Tasks - before writing code. Use for "plan a feature", "new increment", "let's build X".
sw-increment
Plan work as a SpecWeave increment - one spec.md with Problem, Scope, numbered ACs, Approach and Tasks - before writing code. Use for "plan a feature", "new increment", "let's build X".
done
Close a SpecWeave increment - every task done or skipped, specweave verify green, a fresh-context review, then specweave complete. Use for "close it", "we are done", "finish up".