(builtin-toolkit)(builtin-toolkit opt)The enabled builtins as a toolnexus.tool toolkit.
The enabled builtins as a `toolnexus.tool` toolkit.
The ten §4A builtins, in the order of the SPEC table.
The ten §4A builtins, in the order of the SPEC table.
§4A: timeout?:number(ms, default 60000).
§4A: `timeout?:number(ms, default 60000)`.
(enabled-builtins opt)The builtin tools this config exposes. A whole-source-off SHORT-CIRCUITS:
the tools map is never consulted (§0.11).
builtins.tools with a true value. REPORTED.Two sections of SPEC.md specify DIFFERENT FUNCTIONS for the same input, while each claims to be the other:
§4A + §4 'Assembly order & builtin toggle' (and §0.11 itself) — 'a name→bool map applied on the ALL-ON BASELINE: a tool mapped to false is dropped, true (or absent) stays on.' ⇒ {:tools {:bash true}} = ALL TEN.
§3 S2 (skills filter) and §2 Gap 7 (MCP per-server filter), both of which describe themselves as 'identical to ... builtins §4A' — 'nil/empty ⇒ all; ≥1 true ⇒ ALLOWLIST (only true-mapped names)' ⇒ {:tools {:bash true}} = BASH ONLY.
This implements the §4A wording — the section that OWNS builtins, states it twice, and which §0.11 restates. Spike s22 made the same call. Whichever way it is resolved, §3-S2 and §2-Gap-7 must stop claiming they are identical to §4A unless they are. Six ports cannot be byte-identical against a spec that specifies two behaviours for one input.
The builtin tools this config exposes. A whole-source-off SHORT-CIRCUITS:
the `tools` map is never consulted (§0.11).
===================================================================
KNOWN SPEC DEFECT — `builtins.tools` with a `true` value. REPORTED.
===================================================================
Two sections of SPEC.md specify DIFFERENT FUNCTIONS for the same input,
while each claims to be the other:
§4A + §4 'Assembly order & builtin toggle' (and §0.11 itself) —
'a name→bool map applied on the ALL-ON BASELINE: a tool mapped to false
is dropped, true (or absent) stays on.'
⇒ {:tools {:bash true}} = ALL TEN.
§3 S2 (skills filter) and §2 Gap 7 (MCP per-server filter), both of which
describe themselves as 'identical to ... builtins §4A' —
'nil/empty ⇒ all; ≥1 true ⇒ ALLOWLIST (only true-mapped names)'
⇒ {:tools {:bash true}} = BASH ONLY.
This implements the §4A wording — the section that OWNS builtins, states it
twice, and which §0.11 restates. Spike s22 made the same call. Whichever way
it is resolved, §3-S2 and §2-Gap-7 must stop claiming they are identical to
§4A unless they are. Six ports cannot be byte-identical against a spec that
specifies two behaviours for one input.(glob-match? pattern path)** crosses separators, */? do not — so a recursive sweep is **/*.md.
`**` crosses separators, `*`/`?` do not — so a recursive sweep is `**/*.md`.
(render-questions questions)§4A, byte-identical across ports: each question's text in order, with
" (options: a, b, c)" appended when it has non-empty options, joined by
"\n", no trailing newline. header is NOT rendered — it stays in
data.questions.
§4A, byte-identical across ports: each question's text in order, with " (options: a, b, c)" appended when it has non-empty options, joined by "\n", no trailing newline. `header` is NOT rendered — it stays in `data.questions`.
(source-on? opt)Whole-source gate. builtins:false | {disabled:true} | {enabled:false}
turn the source off, with MCP precedence: disabled:true wins, else
enabled:false disables, else on. Absent ⇒ ON (the default).
Whole-source gate. `builtins:false` | `{disabled:true}` | `{enabled:false}`
turn the source off, with MCP precedence: `disabled:true` wins, else
`enabled:false` disables, else on. Absent ⇒ ON (the default).UTF-8 byte length of s — koine.text/utf8-length.
The hand-rolled version here was WRONG on both hosts (8 and 5 bytes for a 6-byte answer — it folded code units, not code points) and was fixed locally before koine 0.11.0 lifted the correct fold into the seam. Delegating keeps one implementation; the regression test in builtin_test stays here, guarding the call site rather than trusting the seam blindly.
UTF-8 byte length of `s` — `koine.text/utf8-length`. The hand-rolled version here was WRONG on both hosts (8 and 5 bytes for a 6-byte answer — it folded code units, not code points) and was fixed locally before koine 0.11.0 lifted the correct fold into the seam. Delegating keeps one implementation; the regression test in builtin_test stays here, guarding the call site rather than trusting the seam blindly.
cljdoc builds & hosts documentation for Clojure/Script libraries
| Ctrl+k | Jump to recent docs |
| ← | Move to previous article |
| → | Move to next article |
| Ctrl+/ | Jump to the search field |