| name | agentic-maturity |
| description | Place the workspace owner on the Agentic Maturity (AM) model and give
them a prescribed next step. Seeds objective signals from `aios analyze` (local
session logs) when available, confirms with a short interview, applies the
verification cap, writes the placement + journey to .claude/memory/MATURITY.md, and
prescribes the highest-leverage patterns to practice next. Use when the user says
"assess my agentic maturity", "level me up", "what should I learn next", "how good am
I with AI", "rate my AI workflow", or on a periodic check-in to track progression.
|
| kind | skill |
| version | 1.0.0 |
| triggers | ["assess my agentic maturity","agentic maturity","level me up","what should I learn next","how good am I with AI","rate my AI workflow","am I getting better at agentic engineering","my AEM level","my AM level"] |
Agentic maturity self-assessment + journey
Place the owner on the AM Spine (L1–L5) and score the five Axes (0–4), then
write a durable placement and prescribe the next patterns to practice. The model and
rubric are canonical in agentic-engineering-maturity/ (root); this skill reads the
local copies: the scoring bands in individual.rubric.json and the pattern catalogue
in curriculum.md (both in this skill folder).
Core rule (do not violate): the Spine level is capped at L3 if the Verification
axis scores ≤ 1. No real agentic maturity without verification — say so plainly if it
applies.
Asking (chat UI): ask in plain chat messages — do NOT use the AskUserQuestion
tool (it can't render in the cockpit). Ask the confirmation questions as one short
numbered batch and invite a free-form reply.
Step 1 — Seed objective signals (best effort)
Run the local analyzer to get signal-based scores from the owner's real session logs:
npm run aios -- analyze --since 30d --json
If it returns a placement, use its placement.spine, placement.axes, and
placement.weakest as the starting hypothesis. If no logs are found (new machine,
other tools), skip to the interview and place from answers alone — say which you used.
Step 2 — Confirm with a short interview
Signals are a proxy; confirm and adjust. Ask these as one batch (answer in any order):
- Spine — which is most true under pressure (not your best day)?
read the five
spine[].placement lines from individual.rubric.json.
- Verification — when an agent finishes, how do you confirm it's correct?
- Context — how do you manage what the model sees across a session?
- Autonomy — how do you decide how much an agent does on its own?
- Learning — when an agent makes a mistake, what happens next time?
- Cost/governance — how aware are you of token cost and tier/permission discipline?
- Delegation — roughly what % of your work is delegated-and-verified to agents?
Map each answer to a 0/2/4 band using the axes[].bands in the rubric. Reconcile with
the signal hypothesis; when they disagree, trust the interview but note the gap.
Step 3 — Place + apply the cap
- Spine = the owner's reliable default mode, cross-checked against the signal level.
- Apply the verification cap: if Verification ≤ 1, hold the Spine at L3 (or below).
- Identify the weakest axis — it drives the prescription.
Step 4 — Prescribe the next step
Look up the matching entry in patternMap (by spine, and weakest-axis when present),
then name the patterns from patternTitles. Read the matching section of
curriculum.md and give the owner 2–3 concrete patterns to practise next, each with
the single first action they should take this week. Lead with the highest-priority one
(verification-first entries are marked "priority": "highest").
Step 5 — Write the durable record
Update .claude/memory/MATURITY.md:
- Set the Current placement block (date, Spine level + name, the five axis scores,
weakest axis, whether the cap applied, and whether signals or interview drove it).
- Append a one-line History row so progression is visible over time.
- Set Active journey to the prescribed module + patterns.
Then tell the owner their placement in one tight paragraph, the one rule if the cap bit,
and their first action. Offer to re-check in a few weeks (progression shows in History).
Quality bar
Before finalising, self-check against .claude/rubrics/agentic-maturity.md: the cap was
applied correctly, the prescription targets the weakest axis, and MATURITY.md was
actually written. Revise until it passes.