Your named specialists, and your own assistant's three faces.
You run your life through specialists you call by name rather than one general helper.
As of September 10, 08:02 PM
What the status words mean
proven · done, not independently checked · partly done · not started · blocked · parked0%not started
19 steps: 0 proven · 0 partly done · 18 untouched · 0 blocked · 1 parked
agent runs · not recordedtokens used · not recordedreview notes · not recordeddollars charged · not recorded
The plan as first written
Every agent that runs is one Nick owns and authorised, does a job no other agent does, follows his current rulings, and leaves a record naming which specialist and which model actually ran. Skippy, Gracie and Neeko stop being three separate things and become one assistant with three faces, each with a manual Nick can hand to anyone. In his own words this morning: *"we only dispatch agents we own and authorized so the whole plan should be to develop the roster to reflect our needs"* and *"they all need to be under one core file as they all are the same agent just with a different face and capabilities listed."*
Steps
| Id | Plain words | Percent | State | Proof | What is left |
|---|---|---|---|---|---|
| STEP 1 | Settle why Larry was reported missing, fix the header that hides his routing rule, and prove him by calling him | 0% | not started | Python3 projects/ops/life-os/REGROUP-2026-09-08/plans/AGENTS/check-larry-frontmatter.py seen red then green, plus evidence/step1-larry-two-readings.txt carrying both session listings with their dates and surfaces and Larry's kept reply | The step |
| STEP 2 | Write down every element the team page draws, then build the check that measures them and prove the check can go red | 0% | not started | Node projects/ops/artifacts/agent-org-chart/orgchart-fidelity-check.mjs --selftest prints 0 · 0 and the sabotage run goes red naming EVERY compared property, both saved to evidence/step2-orgchart-fidelity.txt | The step |
| STEP 2b | Two specialists show on the team page as lowercase code names — put the names you use on them, and prove the page did not shift | 0% | not started | The four-part publish gate: the fidelity check printing mismatched properties: 0 · unmeasured anchors: 0 once for each of the four viewport×theme pairs into evidence/step2-orgchart-fidelity.txt, the labelled side-by-side evidence/step2-orgchart-side-by-side.png, Sienna's verdict, the verifier's verdict — plus evidence/step2b-directory-names.json carrying a disposition for every node and Benito's kept reply | The step |
| STEP 3 | One list of every agent we own and authorise — what it is for, who calls it, what it may touch | 0% | not started | Python3 projects/ops/agents/build_agents.py --validate-only reports zero retirement, and evidence/step3-roster-coverage.txt lists all 27 specialist files against their four-field rows, the six built-in helpers as AWAITING NICK with their counts, and the three not named names as refusable | The step |
| STEP 4 | Nothing gets dispatched under a name we did not authorise | 0% | not started | Node projects/ops/skippy-jobs/lib/test-dispatch-gate.mjs passes with five cases transcribed into evidence/step4-gate-four-cases.txt: made-up name refused, roster name allowed, dormant name refused as dormant, awaiting-Nick name allowed and logged to a named reader, unreadable roster failing open with the failure delivered | The step |
| STEP 5 | After the fact, we can always say which specialist ran and on what | 0% | not started | Evidence/step5-dispatch-month-table.txt carries the last month re-read as one table with its population and its date, plus a new dispatch's own row showing both fields; node projects/ops/skippy-jobs/_test-dispatch-compliance-log.mjs passes beside it as a regression check | The step |
| STEP 6 | A file that reads correctly is not an agent that works — call every one of them that is not parked | 0% | parked | Evidence/step6-invocation-table.md holds 27 rows — 21 kept replies and 6 dated dormancy reasons taken from the roster's own flags; node projects/ops/skippy-jobs/_test-agent-builder.mjs passes beside it as a regression check | The step |
| STEP 7 | Skippy, Gracie and Neeko become one assistant with three faces, each face listing only its own capabilities | 0% | not started | The four files are opened and read by the checker, and evidence/step7-before-after-capabilities.md shows every rule and capability from the three original sources present exactly once across them; python3 projects/ops/life-os/audits/A6/verify_agent_descriptions.py passes beside it as a generator regression check | The step |
| STEP 8 | Three user manuals anyone can read — what each face is for, what it does, what it refuses | 0% | not started | The three manuals are opened by the checker, and evidence/step8-manual-cold-read.md holds a cold reader's capability list per manual set against that face file's own list, with the plain-English verdict; python3 projects/ops/life-os/audits/A6/verify_instruction_checks.py passes beside it as a regression check | The step |
| STEP 9 | Prove the three faces are real by using them, not by reading their files | 0% | not started | Evidence/step9-six-exchanges.md holds six kept exchanges — three in-scope answers and three out-of-scope refusals, each naming the face and the channel; node projects/ops/skippy-jobs/_test-dispatch-compliance-log.mjs passes beside it as a regression check | The step |
| STEP 10 | Your seven approved rule changes stop being drafts and start being what agents follow | 0% | not started | Evidence/step10-seven-readbacks.txt holds the changed passage read back from each of the seven live files themselves with the before-hash re-checked at the moment of applying; python3 projects/ops/life-os/audits/A6/scorecard.py passes beside it as a regression check | Waiting on Nick: seven approval taps — each of the seven changes is named in one plain line in FOR NICK |
| STEP 11 | The six marketing specialists sit dormant under the manager role you will build, and hold nothing up | 0% | not started | Evidence/step11-six-dormant-refusals.txt holds the dispatch gate's own refusal text for each of the six, captured before any dispatch happens, plus the placeholder row naming Nick as owner; node projects/ops/skippy-jobs/_test-agent-builder.mjs passes beside it as a regression check | The step |
| STEP 12 | Four possible ways to combine near-identical specialists, attacked first, written up on one page you can answer from | 0% | not started | Evidence/step12-merge-options-one-page.md exists, names all four merges with what each combines, what it saves and what could be lost, and every claim on it has been re-derived from the six specialist files by the checker | The step |
| STEP 12b | Whatever you say about the merges is what happens — including nothing | 0% | not started | Python3 projects/ops/agents/publish_agent_list.py --json passes and the roster and the directory match his recorded answer exactly, with the before-and-after roster diff kept beside his verbatim words | Waiting on Nick: which merges — all, named, or none |
| STEP 13 | Progress notes stop landing on cards they have nothing to do with | 0% | not started | Node projects/ops/skippy-jobs/_test-zion17-board-and-unified-update.mjs passes with the broadcast reproduced red before the change and refused green after, both transcribed into evidence/step13-board-red-green.txt with the live card read back | The step |
| STEP 14 | A leftover copy sits in the live folder and three links point at nothing — clear both, and never delete anything version control cannot get back | 0% | not started | Node projects/ops/skippy-jobs/_test-status-regen-sp6.mjs passes as a regression check, and evidence/step14-pointer-readbacks.txt records the stray copy's tracked status before removal, its removal commit, the three repointed references each opened and resolving, and the older .bak recorded as already absent | The step |
| STEP 15 | The team's health number gets re-measured every week inside Larry's weekly review, instead of remembered | 0% | not started | Evidence/step15-weekly-run-readback.txt holds one natural run's output read from where it lands, the registration receipt inside Larry's weekly review, the named reader, and the number with its population and date; node projects/ops/skippy-jobs/_test-zion-guards.mjs passes beside it as a regression check | The step |
| STEP 16 | A half-finished change can no longer spend one of your approvals | 0% | not started | Node projects/ops/skippy-jobs/_test-check-root-ticket-gate.mjs passes with all five cases transcribed into evidence/step16-approval-boundary-cases.txt: one valid change succeeding once, and a stale hash, a wrong path, a replay and two competing changes each failing with nothing written | Waiting on Nick: the security click that opens end-phase security work |
| STEP 17 | The checks that keep agents honest test your current decisions, not last month's | 0% | not started | Node projects/ops/skippy-jobs/_test-regroup-skill.mjs and python3 projects/ops/life-os/audits/A6/duplicate_rule_detector.py --selftest both pass, with the removed-duty fixture and the deliberately stale copy kept in evidence/step17-instruction-check-fixtures.txt alongside their red runs, and every repeated rule group carrying an owner and a disposition | The step |
Done
0 of 19 steps proven
Left
19 of 19 steps not yet proven
Blocked on
- Nothing currently blocked.
Formal remaining plan
| # | Step | Needs Nick |
|---|---|---|
| 1 | Settle why Larry was reported missing, fix the header that hides his routing rule, and prove him by calling him | No |
| 2 | Write down every element the team page draws, then build the check that measures them and prove the check can go red | No |
| 3 | Two specialists show on the team page as lowercase code names — put the names you use on them, and prove the page did not shift | No |
| 4 | One list of every agent we own and authorise — what it is for, who calls it, what it may touch | No |
| 5 | Nothing gets dispatched under a name we did not authorise | No |
| 6 | After the fact, we can always say which specialist ran and on what | No |
| 7 | A file that reads correctly is not an agent that works — call every one of them that is not parked | No |
| 8 | Skippy, Gracie and Neeko become one assistant with three faces, each face listing only its own capabilities | No |
| 9 | Three user manuals anyone can read — what each face is for, what it does, what it refuses | No |
| 10 | Prove the three faces are real by using them, not by reading their files | No |
| 11 | Your seven approved rule changes stop being drafts and start being what agents follow | Yes |
| 12 | The six marketing specialists sit dormant under the manager role you will build, and hold nothing up | No |
| 13 | Four possible ways to combine near-identical specialists, attacked first, written up on one page you can answer from | No |
| 14 | Whatever you say about the merges is what happens — including nothing | Yes |
| 15 | Progress notes stop landing on cards they have nothing to do with | No |
| 16 | A leftover copy sits in the live folder and three links point at nothing — clear both, and never delete anything version control cannot get back | No |
| 17 | The team's health number gets re-measured every week inside Larry's weekly review, instead of remembered | No |
| 18 | A half-finished change can no longer spend one of your approvals | Yes |
| 19 | The checks that keep agents honest test your current decisions, not last month's | No |
Decisions this lane is waiting on
- None.