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 agentmods add skills/pascalebeier/hitkeep/hitkeep-developmentnpx skills add PascaleBeier/hitkeep --skill hitkeep-developmentgit clone --depth 1 https://github.com/PascaleBeier/hitkeepWhat 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 | $0.00066 | $0.01203 |
| Opus 5 | $0.00033 | $0.00602 |
| Sonnet 5 | $0.00013 | $0.00241 |
| Haiku 4.5 | $0.00007 | $0.00120 |
Grade A, and why
hitkeep-development 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 2d 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 — 59 lines — stays where its author put it; the contents beside it link to each section on GitHub.
HitKeep Development
Treat AGENTS.md as policy and hk as workflow truth. Never copy commands, build tags, environment defaults, variants, ports, tool versions, or QA gates into another instruction surface.
Connect
Use the central local HitKeep developer MCP for every operation it exposes. It is a stateless server-catalog stdio adapter over the same worktree-confined services as the CLI and is separate from the production analytics /mcp endpoint. A configured or enabled host entry is not enough; confirm that the relevant callable hk_* tools are exposed and make a read-only call before relying on MCP.
Follow this connection order:
- Inspect the host's callable tools and select the relevant
hk_*operation instead of translating it into shell. - Start with a read-only status or planning call and verify that its workspace ID matches the configured fallback or explicit workspace selector. Pass the optional
workspaceinput only when selecting a different catalogued worktree; it accepts a catalogued workspace ID or absolute path. - If tools are absent or fail, inspect host MCP health and compare the configured server name, executable, and arguments with
./hk mcp manifest --output json. Report whether registration, startup, workspace routing, or task reload is blocking MCP. Existing tasks may need a host reload before a corrected registration becomes callable. - Obtain explicit user approval before invoking any equivalent structured CLI action. After approval, discover the command and flags through
./hk catalog commands --output json, request--output json, and consumeschema_version,status,workspace_id,data, anderror. Stop on an unknown schema version instead of guessing.
Reserve direct ./hk use for MCP bootstrap or repair, an explicitly approved fallback, and operations intentionally absent from MCP such as formatter or Go migration rewrites. Never guess a workspace or silently edit client-owned global configuration.
Use the typed MCP surface rather than translating an action into shell yourself:
What ships with it
5 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.
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.
- 2d ago First seen · 59 lines · 66 tokens per session scan A c51d8e18262a
hitkeep-development is a skill published in the GitHub repository PascaleBeier/hitkeep (85 stars, last pushed 3d ago), licensed MIT. It adds 66 tokens to every session and 1,203 once invoked, about $0.0003 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.
Other skills, from other repositories
papercut
Capture or triage small Neva repository and engineering-harness frictions without derailing feature work.
skip-test-with-issue
Skip a flaky or broken test with proper issue tracking. Use when asked to skip a test, disable a test, or mark a test as flaky.
kanban-md
Manage project tasks using kanban-md, a file-based kanban board CLI. Use when the user mentions tasks, kanban, board, backlog, sprint, project management, work items, priorities, blockers, or wants to track, create, list, move, edit, or delete tasks. Also use for standup, status update, sprint planning, triage, or…
kanban-based-development
Autonomous, parallel-safe development workflow using kanban-md. Use when the user asks to work through tasks, do kanban-based development, or when multiple agents need to coordinate work on the same codebase. Optimized for explicit handoffs and a "defer to user" protocol when human intervention is required.
triage-issues
Триаж открытых ишью ivanarama/onebase — классификация, воспроизведение, план фикса. Очевидные дефекты помечает ready-fix (уходят в автофикс без человека), для остального пишет оценку целесообразности и решение оставляет человеку. Этап конвейера сопровождения, запускается по расписанию через PromptPilot.
tail-issues
Заведение заявок по хвосту ревью ivanarama/onebase — находки, которые ревью пометило [заявка] и которые пережили мерж PR. Этап конвейера сопровождения; расписания в PromptPilot пока нет, зовут вручную.