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.
npx skills add YSAA1/harness-workflow --skill sweepgit clone --depth 1 https://github.com/YSAA1/harness-workflowWrote 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/ysaa1/harness-workflow/sweep)<a href="https://agentmods.dev/skills/ysaa1/harness-workflow/sweep"><img src="https://agentmods.dev/badge/skills/ysaa1/harness-workflow/sweep/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/ysaa1/harness-workflow/sweep"><img src="https://agentmods.dev/badge/skills/ysaa1/harness-workflow/sweep.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00076 | $0.01647 |
| Opus 5.5 | $0.00030 | $0.00659 |
| Sonnet 5.5 | $0.00015 | $0.00329 |
| Haiku 4.5 | $0.00008 | $0.00165 |
Grade A, and why
sweep 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 1mo 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 — 54 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Sweep
Repo-wide reconciliation: bring task artifacts, recovery-surface state, and repository leftovers into a single ledger. Detection relies on the checklist, conviction relies on evidence, and ambiguous items go to human adjudication.
This skill is the on-demand entry for low-frequency global reconciliation; it does not do task close-out (close-out goes to cleanup). It complements the retirement contract (the AGENTS.md write-side discipline): when leak-plugging at the source operates normally, this skill most often ends with zero modifications. The target project has no health-check script gate — the inline reconciliation checklist in this file is the equivalent compensation, relying on no script outside the repository.
Process
- Probe the recovery surface (measured, not assumed): inventory the actual file set of
.harness/(commonly the three files, but heterogeneous forms such as decisions/progress also occur). Present → root set = outbound links of all state files + persistent entries (AGENTS.md, root README, docs/README-like files). No.harness/, or the project uses a tracker/existing backend → degrade to scanning only task-artifact orphans and untracked leftovers, explicitly declaring "the.harnessstate surface is missing" (for tracker/existing the recovery surface is the tracker itself, which is not a case of missing), and do not fabricate state criteria (a third-party project may never have created.harness). When state is unparseable, skip the state-contradiction check and say so; do not draw orphan conclusions. - Reconciliation checklist (the manual version sharing the same criteria as the recovery-surface lint of the harness-workflow method repo; runs in target projects that lack the script):
- State contradictions: whether state's Status matches the corresponding work_index row; whether the track pointed to by state is registered.
- Existence: whether the primary artifact of an active/blocked row exists (in the file system or within the trees registered by
git worktree list; a hit in either counts as existing). - Reachability: for each task artifact (plans/specs/reports),
grep -Fits basename across all files in the root set; zero mentions = orphan candidate; basename matching guards against bare-filename false positives. - untracked (unique to sweep, not checked by the lint): listed via
git status --porcelain, already-ignored entries removed viagit check-ignore -v, grouped by directory/theme.
- Six-bucket classification:
- ① Task-artifact orphans: convict tracked ones via triple-evidence conviction — zero mentions in the root set +
git log --followorigin tracing + same-directory cross-references; spec/plan paired artifacts (shared slug) are adjudicated as a group, preventing deleting a spec while leaving its plan; reports-type items with "zero references + README declared non-authoritative" can be charged directly. Self-registered-deliverable guard: research logs (header carries Track self-registration, terminal state written) and review reports (landed indocs/reviews/, header carries an ownership declaration) are self-registered deliverables — zero references does not convict them directly; after triple-evidence conviction they are downgraded to needs-review for human adjudication, never auto-deleted. - ② Zombie blocked rows: not only literally blocked — paused / awaiting-user-confirmation / blocked-by-policy and other equivalent forms are all filed, listing the freeze starting point and duration.
- ③ Half-flipped rows: any row whose status conflicts with actual progress/timeline qualifies; no online verification needed. Two forms: an active/blocked row superseded by a later task but not flipped (e.g., the GPU/directory the old row occupied has been taken over by a new task); a row already flipped to complete but with the four retirement steps unfinished — the plan/Spec derivative documents it links out to still remain under docs/ (a dead pointer on a complete row is the normal state of an already-retired row; the artifact being a live file does not count either).
- ④ untracked leftovers: never deleted; only list evidence + an either/or outlet (add a commit or add a .gitignore entry); directories with tracked precedent (e.g., artifacts/) default to commit candidates; directories containing registered worktrees are listed separately as high-risk —
git worktree removemust happen before any disposition can be discussed; never directly ignore/delete. - ⑤ Archive directories (docs/archive/ and the like): only inventoried and marked; their contents always go to needs-review for human adjudication, never into the deletable bucket.
- ⑥ Recovery-surface files overstepping their remit: state exceeding its snapshot role, decisions/lessons double bookkeeping → distill into lessons then truncate/merge; genealogy-style entries compressed into a single lineage table; among multiple copies of the same rule, keep only the one in lessons.
- ① Task-artifact orphans: convict tracked ones via triple-evidence conviction — zero mentions in the root set +
- Disposition: mechanical items (flipping status, deleting convicted orphans) are fixed automatically; ambiguous items (abandoned vs paused, where unfinished items go) list evidence for user adjudication — no adjudication, no deletion (conservative stop).
- Four retirement steps (same as the AGENTS.md write-side discipline, completed within one commit; historical complete-but-not-retired residue found in ③ is likewise made up): flip the row → lessons inheritance (distill first) → delete this track's plan/Spec documents, including unapproved and abandoned ones (precondition = already committed to the main tree and lessons inheritance done; in-degree guard = only delete documents whose sole inbound edge comes from this retiring track; for multi-edge documents flip the pointer instead of deleting; ADRs are never deleted, add superseded-by) → state syncs by flipping Status.
- Report: with no drift, end with zero modifications.
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.
- 1mo ago Changed · +4 tokens per session ad55133a164b
- 1mo ago First seen · 54 lines · 72 tokens per session scan A ded184c146f1
sweep is a skill published in the GitHub repository YSAA1/harness-workflow (3 stars, last pushed 16d ago), licensed MIT. It adds 76 tokens to every session and 1,647 once invoked, about $0.0003 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 2026-09-21.
Other skills, from other repositories
comet-github-idea-to-issue
A workflow for turning a local idea, bug observation, or improvement proposal into a focused GitHub Issue draft. GitHub Issues are shared records for bugs, features, questions, and maintenance work.
merge-review
Reviews pending fleet worktree merges before they're accepted. Reads the merge-check queue, detects file-level conflicts between branches, proposes a safe merge order, and surfaces reconciliation plans for overlapping changes.
ijfw-ship
Use when the user says 'ship it', 'ship this', 'release', 'publish', 'launch', 'deploy', 'go live', 'wrap this up', 'time to ship', or invokes '/ijfw-ship'. Domain-aware release — software (test → tag → push → publish), book (final edit → format → KDP / agent / Substack), campaign (review → schedule → launch →…
jira
Jira integration — view issues, generate branches, create tickets, transition status. Use when: user mentions Jira ticket key (XX-123), says /jira, wants to create branch from ticket, create a new ticket, or update Jira status. Not for: GitHub issues (use issue-analyze).
lov-rename-project
A project-renaming workflow that updates a product or repository name across source files, configuration, and documentation. It produces a reviewable plan and preserves compatibility-sensitive names and references where needed.
lov-gh-tidy
Interactive GitHub repo hygiene skill. Lists all open issues, PRs, stale branches, and orphan labels, shows a summary of each with analysis, then asks the user how to handle each item (close, merge, comment, delete, keep). Executes all chosen actions via gh CLI. Use when the user says "清理 GitHub", "tidy repo", "clean…