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 commands/missingpackage/nightshift/weekly-maintenancegit clone --depth 1 https://github.com/MissingPackage/nightshiftWhat 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 | $0.00022 | $0.00637 |
| Opus 5 | $0.00011 | $0.00318 |
| Sonnet 5 | $0.00004 | $0.00127 |
| Haiku 4.5 | $0.00002 | $0.00064 |
Grade A, and why
weekly-maintenance 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 yesterday.
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.
What it actually says
Weekly maintenance of the harness (ruling C10, 2026-08-12: runs scheduled via tools/install-schedules.sh; the old "100% PI time" ritual was never executed — 0 out of 3 weeks. The agent does the checks; only the decisions reach the user). Write the report itself in the user's own language — user surface, "match the user's language" rule; only these instructions are English. Plain language, titles not IDs. Order:
- Cross-project docket triage. Every
.harness/goals/*/docket.mdunder ~/Projects (skipSTATUS: PAUSEDgoals): open entries with age in days, oldest first. For each a proposal: decide now / can wait / kill — with the why in ≤15 words. The proposals feed the decision batch of the next coffee report. - Agentic adversarial audit (replaces the PI's spot-audit — ruling C10: he does not
read diffs, by choice). Pick ONE PASS-verified slice of the week at random
(
git log --oneline --since="1 week ago" | shuf -n1, no cherry-picking) and launch an independent adversarial agent (adversarial-reviewer) with a mandate to REFUTE the PASS: diff vs done-when vs evidence. If the audit contradicts the verifier: harness incident — tighten the verifier the same day, and at the top of the next coffee report. - Hook health (10s each).
bash tests/run.shin the harness repo; plus the live type-tests: "fatto?" fixture → firefight-catch responds; push fixture → push-guard decides per policy; session-anchor on the repo → emits HANDOFF §1. A silent hook = incident. - Memory and surface hygiene. HANDOFF over 80 lines per touched project; goals with all phases done but not archived; auto-memory MEMORY.md with dead entries; GLOSSARY.md: entries whose referent no longer exists (grep the term in code/docs: zero hits = removal candidate) and terms coined in the week's digests that are NOT in the glossary; pruning skill sediment: references to files or IDs a fresh reader cannot resolve — an installed skill must stand on its own. Propose the pruning diff, do not apply it.
- Eval drift (if evals/ exists). Run and compare with the latest RESULTS: a dropped section is an incident; one stuck at 100% for 3+ weeks calls for harder scenarios.
Report ≤ 50 lines: one section per point, each closed by "nessuna azione" (no action) or by the exact decision (which flows into the next morning's frontier batch).
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.
- yesterday First seen · 39 lines · 22 tokens per session scan A 55488230e790
weekly-maintenance is a command published in the GitHub repository MissingPackage/nightshift (2 stars, last pushed 15d ago), licensed MIT. It adds 22 tokens to every session and 637 once invoked, about $0.0001 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 2026-08-31.
Other commands, from other repositories
research
Research a technical or product question.
safe-build
Build the application for development or production.
update
Use when: search results seem noisy, after bulk imports, or for periodic memory/registry maintenance.
plan-workflow
Generate a skill chain plan for a high-level goal by analyzing the skills registry for input/output compatibility.
music-suno-prompt
Grounded Suno prompt synthesis from local knowledge corpus + persona canon + label canon. No vibes-prompting.
lrn
Execute the /vibeguard:learn command. $ARGUMENTS.