agentleFS
iwesysiwesys/IWE/.claude/skills/verify-hypotheses/SKILL.md

verify-hypotheses

Skill0 starsChanged 25 days agoMITView on GitHub

Verification of the hypotheses log (hypotheses-log.md) on demand, outside

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not issue a verdict without checking against real evidence — "it seemed confirmed" does not count.

    line 84
  2. Distinctive:

    Do not edit the original hypothesis entry when issuing a verdict — only a new entry with a reference is allowed.

    line 85
  3. Distinctive:

    Do not issue "refuted" instead of "not applicable" when the criterion's precondition was not met.

    line 86
  4. Distinctive:

    Do not leave a verdict without a chosen follow-up action — a hanging verdict does not close the cycle.

    line 87
  5. Distinctive:

    Do not run this skill automatically on every session — only on an explicit pilot request (the weekly rhythm is covered by step 6a of …

    line 88

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/verify-hypotheses

Scope: read hypotheses-log.md, find entries whose verification date has arrived, check them against available evidence, propose a verdict, write the pilot-confirmed verdict as a new entry. Not in scope: does not create new hypotheses (that is Note-Review), does not run verification automatically on week-close (that is step 6a in .claude/skills/week-close/SKILL.md — this skill duplicates the same algorithm for use outside the weekly rhythm), does not edit original entries. Role: no dedicated role owner — used by whoever is currently running week-close (Strategist, R1). See open ArchGate question WP-496 F4.

When to use

  • The pilot says "verify this week's hypotheses" or "verify hypotheses" outside the normal Week Close rhythm.
  • There is reason to believe a specific hypothesis has already resolved (the pilot observed the evidence before the scheduled verification date) and wants to check early.
  • Diagnostics: how many hypotheses are currently "pending verification" and what are their dates.

Preconditions

  1. WP Gate precondition. The task is linked to WP-496 (Hypotheses log, LPF feedback regulation), phase 3.
  2. Log exists. {{GOVERNANCE_REPO}}/current/hypotheses-log.md — if the file does not exist, notify the pilot and stop (the log has not been created yet or was created in a different location).

Algorithm

Step 1 — Read the log and filter

Input: {{GOVERNANCE_REPO}}/current/hypotheses-log.md Action: read all entries. Select those where Status: pending verification AND Verification date ≤ today. Skip entries with status draft (not yet confirmed by the pilot in Note-Review) — they are outside the verification cycle. Skip entries that already have a verdict (confirmed / refuted / partially confirmed / not applicable). Output: list of IDs (H-NNN) with a due verification date, ready for review. If the list is empty — report "No hypotheses with a due verification date" and stop (not an error; this is a normal result).

Step 2 — Check each entry against available evidence

Input: list of IDs from Step 1 Action: for each entry, read the falsification criterion and check it against available sources — commits (git log for the affected repositories), domain_event (if the criterion concerns a measurable platform metric), infrastructure logs, facts from the nearest WeekReport/DayPlan. If the criterion requires data that is not available in the accessible sources — explicitly mark it "insufficient data for a verdict" and do not guess. Output: for each entry — a draft verdict (confirmed / refuted / partially confirmed / not applicable) with justification (which facts were checked and where they came from).

"Not applicable" rule: if the precondition of the criterion was physically not met (for example, the dependent artifact the hypothesis concerns was never delivered) — the verdict is "not applicable", not "refuted". It is impossible to distinguish hypothesis failure from implementation failure when the criterion's precondition was not satisfied.

Step 3 — Present to the pilot and obtain confirmation

Input: draft verdicts from Step 2 Action: show the pilot a table (ID, brief statement, proposed verdict, justification). Ask for confirmation or correction on each entry. Output: final verdict for each entry, agreed with the pilot.

Step 4 — Record the verdict and propose a follow-up action

Input: final verdicts from Step 3 Action: for each entry — append a new entry to the end of hypotheses-log.md in the format ## Verification H-NNN with a reference to the original entry, the verdict, and the verification date. Do NOT edit the original entry (retroactive editing is prohibited — memory/lpf-hypothesis-log.md). Update the Status field in the original entry to reflect the outcome (this is the only field in the original entry that changes — status only, not content). Then ask: what action follows from this verdict — update confidence for the future / add a step to a checklist / open a work package / record a candidate pattern (Capture-to-Pack). The chosen action is executed outside this skill (terminal exit — goes to WeekPlan/Pack/work package, the consumer is outside the verify-hypotheses perimeter). Output: log updated, each verdict has an associated action (no verdict is left hanging).

Step 5 — Commit

Input: updated hypotheses-log.md Action: commit to the governance repository (same process as other current/ edits). Output: changes saved, hypothesis history preserved.

Bundled resources

None — the algorithm is entirely text-based, with no scripts or templates.

Anti-patterns

  • Do not issue a verdict without checking against real evidence — "it seemed confirmed" does not count.
  • Do not edit the original hypothesis entry when issuing a verdict — only a new entry with a reference is allowed.
  • Do not issue "refuted" instead of "not applicable" when the criterion's precondition was not met.
  • Do not leave a verdict without a chosen follow-up action — a hanging verdict does not close the cycle.
  • Do not run this skill automatically on every session — only on an explicit pilot request (the weekly rhythm is covered by step 6a of week-close).

Verification

bash .claude/skills/skill-creator/scripts/verify-skill.sh verify-hypotheses

Expected result: PASS. Live check — invoke the skill manually after 2026-09-01 (the verification date of the first entry H-001 in hypotheses-log.md), and confirm that the algorithm finds exactly that entry and no others (the log currently contains only H-001).

<!-- USER-SPACE --><!-- /USER-SPACE -->

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.