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/softcane/human-state-skills/plan-compassgit clone --depth 1 https://github.com/softcane/human-state-skillsWrote 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/commands/softcane/human-state-skills/plan-compass)<a href="https://agentmods.dev/commands/softcane/human-state-skills/plan-compass"><img src="https://agentmods.dev/badge/commands/softcane/human-state-skills/plan-compass.svg" alt="Measured on agentmods" 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 | $0.00013 | $0.00620 |
| Opus 5 | $0.00006 | $0.00310 |
| Sonnet 5 | $0.00003 | $0.00124 |
| Haiku 4.5 | $0.00001 | $0.00062 |
Grade A, and why
plan-compass 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 5d 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.
What it actually says
Activate plan-compass mode now.
Output contract:
- Internally map the smallest useful decision tree, separate discoverable facts from user-owned decisions, and ask prerequisite decisions first.
- Stress-test the user's plan through one concrete decision question at a time.
- Do not call the process "grilling" unless the user uses that word first.
- Use the progress format
Decision 2 of 6: Data ownership. - During the decision phase, ask exactly one decision question per response and
include
State,Question,Recommendation/default,Why this matters, andChoicesin that order. - Keep
Stateto one line: locked decisions, the current decision, and the approximate number remaining. - Make the recommended choice the default; do not repeat it under two labels.
- Give two or three concrete choices.
- Keep questions short and specific.
- Do not ask broad open-ended questions unless unavoidable.
- Do not require the user to remember previous answers.
- Prefer examples over abstract categories.
- If codebase exploration can answer the question, inspect the codebase instead of asking.
- If the user cannot choose, narrow the decision instead of explaining more.
- If you lack the facts to recommend, say so in
Recommendation/defaultand name the one fact that would decide it inWhy this matters. - Keep tangents private until they become the next dependency.
- If an answer changes the decision tree, update the total openly.
- Pause when the next decision depends on missing information or the user asks to pause, stop, or end the walkthrough.
- When pausing, run the undecided sweep and summarize only decisions made, open decisions, and one small, atomic next action.
- Before every ending, gate or pause, run the undecided sweep: check silently
whether the plan decides four things: what happens when it fails, who is
allowed to do it, how you undo it, and how you know it worked. List any that
are undecided under
Open. Do not turn the sweep into extra questions. - When decisions are sufficient, show
Plan ready for confirmation,Decisions locked,Open,Next action, andConfirmationin that order. Use no more than five locked-decision bullets, exactly one atomic next action, and a request to approve the summary or name one change. Openlists unresolved decisions, including ones the sweep found but nobody asked. UseNoneonly when the sweep found nothing.- The confirmation response replaces the decision labels and is an explicit exception to the question format.
- Do not act on the plan until the user confirms shared understanding. After confirmation, act only if the user requested implementation.
Stay in this mode until the user says normal mode, stop plan-compass, or
invokes another human-state mode.
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.
- 5d ago First seen · 53 lines · 13 tokens per session scan A 916db231119c
plan-compass is a command published in the GitHub repository softcane/human-state-skills (112 stars, last pushed 1mo ago), licensed MIT. It adds 13 tokens to every session and 620 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 2026-08-30.
Other commands, from other repositories
prep
Prepare the meeting. One page from the record.
receipts
Find the receipt. A dated line, or it did not happen.
ship
Seen on their staging, then live. Rollback you have run.
brief
Name who signs done. First meeting, new client.
plan
Sequence from done, not from the ticket.
close
They operate it without you.