Memory-derived capabilities
P This is the complete inventory of what the v1 assistant's retained knowledge — its memory tree, its knowledge base, its twenty skills, its SQL and its command-line tools — could become in v2: a tool, a skill, an agent, or an article that seeds a memory node. One line per candidate, so the operating-model diagram can be drawn from a list rather than from recollection.
P It is not a claim that anything here is built, scoped or approved. The authoritative definitions of what is designed remain the stage, agent and tool catalogue (machine-readable: agent-catalog.json) and the operation inventory (machine-readable: frequent-operations.json). Where a row already exists in either, it is marked existing and carries that id; nothing here re-invents one. The machine-readable form of this page is memory-capabilities.json.
P What a skill is, when one exists and how a tool differs from a query recipe are settled in Agents Memory §§ 8–19; a new agent exists only for a distinct responsibility or an isolation boundary (Agents § 0), which is why this inventory proposes exactly one. Frequency is quoted only from Operational Method Evidence, with its row named; no count is invented here.
0. Census
F Counted over this checkout on the date of the provenance marker above.
| Count | Count | Count | |||
|---|---|---|---|---|---|
| Rows in total | 119 | tools | 68 | Support root | 10 |
existing (already catalogued) |
39 | skills | 30 | Data & Manipulation (7 domains) | 58 |
candidate (derivable now) |
60 | articles | 20 | Configuration | 12 |
article (seeds a memory node) |
20 | agents | 1 | Development | 9 |
| distinct seed files cited | 515 | Commercial | 4 | ||
| Common and platform | 16 | ||||
| Workspace-only | 10 |
F The seed corpus behind those rows: 275 of the 369 files in memory/reference, 138 of the 166 in memory/feedback, 9 of the 27 in memory/project, all 3 in memory/user, 23 of the 25 knowledge documents, all 20 v1 skills, 23 SQL files, 13 workspace scripts and 6 of the 7 tool packages (the vendored database client aside).
P How to read a row. kind follows the definitions above — a tool has a bounded input and output a script could implement today, a skill is a procedure with judgement points, an agent is a profile. status is existing where the capability is already in the catalogue or in the operation inventory's helper consolidation (the id is given), candidate where a concrete procedure or a runnable script in this checkout makes it derivable now, and article where the material is knowledge rather than a procedure and therefore seeds a memory node. evidence quotes the measured shape where one exists, otherwise the number of memory files that describe the pattern, counted for this page. Duplicated material is one row with several seeds, never one row per file.
1. Support root
2. Data & Manipulation — transfers, families and cargo
3. Data & Manipulation — pricing and billing
4. Data & Manipulation — documents and printing
5. Data & Manipulation — accounts, access and parties
6. Data & Manipulation — health
7. Data & Manipulation — master data and blanks
8. Data & Manipulation — operations recovery
P The one proposed agent sits here. Five operations in the inventory carry the operational-recovery domain — exhausted storage, an orphaned database-link backlog, a broker/channel recovery, a halted process, a frozen replica — and all five are assigned to the transfer investigator today, because their symptom arrives as a failed transfer. Two things separate them from transfer work. The evidence is platform state, not policy state: session and lock views, tablespace and archiver state, queue depth, host processes. And the connector that reads it is a shell on the estate's hosts, which is not read-only — it carries privileged execution. Profile isolation in this wiki is provisioning, not instruction (Agents § 0.2), so a profile that holds privileged host execution should not be the same profile that holds the customer's policy-write grants for the same case. That is an isolation boundary, which is the only ground on which a new agent exists; a skill of the transfer investigator would inherit the grant rather than separate it.
P Agents considered and rejected. An assistant-failure agent — the Beth shape is 14 tickets on one product surface with no denied context of its own, so it is a skill of the transfer investigator. A health-import agent — the health investigator already owns that surface, and import lines are its data. A corpus-migration agent — the v1 migration is a project with an end, and the platform already spawns a summarizer for a missing node (Agents § 0.6). A partner-integration agent — the commercial assessor plus a knowledge domain covers it, and the partner never reaches the estate.
9. Configuration
| id | kind | what it does | seeds | owner | status | evidence |
|---|---|---|---|---|---|---|
configuration.skill.normalization |
skill | Understand and reconcile product sources — Turn a specification into a validated product model, with the baselines and the blocking questions named. | product config new product · product configuration · /configure-product · find the requirement before proposing change | configuration.normalizer |
existing · configuration.skill.normalization |
Configuration changes close in a median 9 days against 1 for a correction. |
configuration.tool.product-validator |
tool | Product-model consistency checker — Check an authored product model against its schema, recount its members and resolve its cross-references. | validate product · product config new product | configuration.normalizer |
existing · configuration.tool.product-validator |
Already an executable in the wiki's own contracts folder; it cannot confirm the source's meaning, only the model's self-consistency. |
configuration.skill.product-build |
skill | Build a product from its specification — Propose the configuration parts a specification implies, agree the scope, then emit the ordered set, the acceptance script and the rollback. | /configure-product · product config new product · product configuration · config key uniqueness map · dev product landscape | configuration.root |
candidate | The v1 twelve-part build is the single largest skill in the workspace; v2 splits it across the five configuration stages, so this row records the seed rather than a new capability. |
helper.config-diff |
tool | Target-aware configuration diff — Diff schema, rows, effective versions and consuming code for one configuration area between two targets. | note incidental config diffs dont normalize · config key uniqueness map · null semantics per config table · config surface liveness test · product configuration | configuration.ipal.builder |
existing · helper.config-diff |
Serves 7 configuration operations in the inventory. |
helper.numbering-preflight |
tool | Non-consuming numbering preflight — Check rows, sequence, grant, synonym, principal and collisions for a numbering configuration without consuming a number. | policy no generation · manual policy no preconsumes sequence collision · seq polno office script superset · policy no composition · policy no office frozen at first transfer · cargo child number shared series collision | configuration.serdica.builder |
existing · helper.numbering-preflight |
6 tickets asked / 14 analyses used; serves the numbering and the number-collision operations. |
configuration.tool.validation-cell-toggle |
tool | Validation threshold relax and revert — Capture the baseline of a validation cell, relax it for one bounded operation and revert it to the captured baseline. | validation threshold temporary relaxation · field validation frameworks · validation config freshness · cfg manual premium control · check abacus layer for validations · abacus validation pipeline · never promote config ahead of evaluator | configuration.ipal.builder |
candidate | 18 tickets asked / 41 analyses used; 4 holders. It is the clearest write-with-teardown in the corpus. |
configuration.tool.label-locate |
tool | Label, message and translation locator — Find which of the three text surfaces owns a displayed string and emit the per-language update for it. | ipal descr link chain · ui i18n vs sr messages · sr messages before lt doc codes · pricing factor dependent display label · insys messages reins 79 | configuration.serdica.builder |
candidate | 12 tickets asked / 27 analyses used; 4 holders; shared with the configuration label sweep. |
configuration.tool.factor-value-add |
tool | Pricing factor and dependent-choice add — Add a factor value, and its dependency key where the factor is dependent, to every product that needs it, from the counterpart's source row. | ipal pricing factor framework · abc cfg pricing factors mapping · graphql prpricingfactorvalues · 2215 home add dependent lov ui gap · pricing factor dependent display label | configuration.ipal.builder |
candidate | 11 tickets asked / 42 analyses used; 4 holders; the flat-versus-dependent trap has its own memory file. |
configuration.tool.tariff-table-check |
tool | Official tariff table check — Check a computed premium against the transcribed official tariff matrix for the product, band by band. | motor tariff 2026 · 4710 abacus factor defaults · 4704 casco covers factors · 4710 electric vehicle tariff band gap · casco no claims discount years based | configuration.source_analyst |
candidate | The 2026 motor matrices are already transcribed as fixed tables; the tariff analyst is a profile denied a database, so a table it can read locally is exactly its instrument. |
configuration.tool.print-doc-register |
tool | Registered document-set change — Add, remove or re-point a document registration for a product and prove the registration is what the print gate reads. | cfg print docs template configuration · pr basic product documents lov · insis print configuration · sr docs chain | configuration.serdica.builder |
candidate | 4 memory files; the diagnostic side is covered by the print probe, the registration side is not. |
configuration.article.config-surfaces |
article | Where product behaviour is configured — Every configuration surface across the three systems that decides how a product behaves, and which key is unique where. | product configuration · product config new product · config key uniqueness map · null semantics per config table · config surface liveness test · abacus schema · abacus engine vs abc access schema | configuration |
article | 7 knowledge/memory files; two of them exist only to stop the pricing engine and its resident schema being confused. |
configuration.article.product-catalogue |
article | The products and what each one is made of — Per-product cover sets, factors, packages and known configuration gaps, product family by product family. | 4704 casco covers factors · cargo products · product 1126 · marsh 3618 packages · lb offers and cover rules · dev product landscape · travel 3618 iv authority ladder · vessel object model 1126 | configuration |
article | Roughly 40 product-specific memory files; 8 are cited here as the shape of the node. |
10. Development
11. Commercial
| id | kind | what it does | seeds | owner | status | evidence |
|---|---|---|---|---|---|---|
commercial.skill.assess-effort |
skill | Assess required work and effort — Investigate what a request would take through read-only evidence and record the range, dependencies, exclusions and uncertainty. | support contract bg · support contract custom en · insis change request | commercial.assessor |
existing · commercial.skill.assess-effort |
22 of 485 tickets are CR candidates; three operations in the inventory route to assessment rather than delivery. |
commercial.tool.effort-comparable |
tool | Comparable-work lookup for an estimate — Find past cases of the same shape and return their recorded effort and outcome as the comparables behind an estimate. | mine tickets classify · mine tickets timing · kb memory vs rediscovery cost · Operational Method Evidence | commercial.assessor |
candidate | The mining scripts that produced the measured catalogue already exist; 185 reliable timings against 230 human estimates. |
commercial.tool.integration-sample-pack |
tool | Partner integration sample pack — Emit a partner-ready request and response set for one product from what the product declares, with the pricing factors listed. | register policy dto · register policy one shot from db · graphql prpricingfactorvalues · /bulsi · 4710 abacus factor defaults · bul si integration partner | commercial.assessor |
candidate | 6 seeds; one live partner engagement drove all of them and the one-shot builder is already written out. |
commercial.article.partner-engagement |
article | How a partner integration is run — What a third-party developer needs, what may be shared, and the discipline of answering a partner's question. | /bulsi · bul si integration partner · bulsi engagement workflow · partner engagement discipline · verify bulsi mechanics before asserting · no modify partner shared content · bulsi 3636 premium rounding 363626001199130 | commercial |
article | 7 memory files and one v1 skill from a single partner engagement. |
12. Common and platform
13. Workspace-only
P Not a runtime capability. These serve the v1 file-based workspace and have no place in the v2 runtime: the case record, the ledger and the connectors replace what they produce. They are listed so the migration does not mistake their absence for an omission, and because two of them (the documentation gate and the offline viewer) are worth keeping as authoring tools for this wiki itself.
Challenges
- C The operational-recovery symptom domain has no owner of record. Its five operations are filed under the transfer investigator, and the argument above says they should not be. This needs a D-row either way, because it decides whether privileged host execution is provisioned to an existing profile or to a new one.
- C A tool owned by one module and bound by another module's skill has no stated precedent. The rating client is Configuration's, but every pricing investigation calls it; the print probe is Data's, but the configuration document-set change needs the same read. Agents Memory § 16 gives one owner per package and shared bindings to one digest, which permits this — but nothing says who answers when the binding breaks for the borrower.
- C Does the twelve-part product build survive as one skill? It is the largest single procedure in the v1 workspace, and its distinctive act is the scope conversation — proposing the parts a specification implies and asking which are in scope. The five configuration stage skills absorb the emission; nothing yet owns the conversation.
- C Agent and office provisioning is catalogued under the access investigator (
data.skill.agent-office) but half of it configures numbering counters and role audiences. Blank ranges (5 tickets asked) and vehicle catalogue entries (2) sit with master data; agent/office setup (35, and 70 hours) now sits with access. Whether the numbering and role half belongs to Configuration S5 rather than to Data is decided on the first ten access cases. - C Is an outbound-text lint a tool or a profile rule? Ten recorded corrections make a rule list, but a check the model may overrule is not deterministic in the sense of Agents Memory § 19. It publishes as a tool only if its output is a typed finding the communicator must answer.
- C The health payment packet leaves the estate. It produces a file a bank consumes, which is not a target write, a message or a document — the write classes in Gating do not obviously contain it.
- C What seeds the past-case index — the archive miner or the migration pass? The five mining scripts produce per-shape aggregates; Agents Memory § 7 describes a distilled per-domain migration. They differ in granularity, and running both would produce two indexes of the same corpus.
- C The four channel clients are not built where the workspace says they are. Source and a Release build exist for three of them, one has no build output at all, and every skill names a publish path none of them occupies. v2 must decide whether it inherits them as packaged binaries under the tool-package contract or reimplements them as connectors.
- C The health knowledge base carries five overlapping editions of one domain guide and the same data model in up to four places. The inventory files it as one article row; the migration has to pick the canonical pair — one platform-generic guide and one customer-values guide — before the node is written.