Headless
15 specifications, 120 requirements. All libraries
0001 Headless layers
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/layers.test.ts: the real tree has no finding of the layers check; suite/layers.test.ts: every file is in one unit, and a file in none is layers/unclassified |
FR-002 | check | bun run verify | suite/layers.test.ts: the real tree has no finding of the layers check; suite/layers.test.ts: imports point one way: below may not import above, and a surface may not import another surface |
FR-003 | check | bun run verify | suite/layers.test.ts: the real tree has no finding of the layers check; suite/layers.test.ts: a file of one unit reaches another through the package's exports, never by a relative path, and a name exports does not name is layers/private |
FR-004 | check | bun run verify | suite/layers.test.ts: the real tree has no finding of the layers check; suite/layers.test.ts: each unit imports only what its row allows: governance Core, logical Core and the terminal's args and commands, the command line Core, the terminal and the system, the adapter Core and the telemetry |
FR-005 | check | bun run verify | suite/layers.test.ts: the real tree has no finding of the layers check; suite/layers.test.ts: the command-line surface may not import the telemetry; only the adapter may |
FR-006 | check | bun run verify | suite/layers.test.ts: a surface is one folder with one row of the table; a folder with no row is layers/unclassified, and its row says what it may import |
FR-007 | check | bun run verify | suite/layers.test.ts: the check holds the findings of the layers and of the trace against the baseline, which only shrinks, and verify runs it |
FR-008 | check | bun run verify | suite/layers.test.ts: the manifest agrees with the folder: the entries, the name, the level, the version and the distribution |
0002 Headless governance
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/governance-action.test.ts: an action has the facts of its kind by default, and defaults can be overridden |
FR-002 | check | bun run verify | suite/governance-action.test.ts: a report only reads, and an action that only reads is a report |
FR-003 | check | bun run verify | suite/governance-action.test.ts: a destructive action must preview and is not idempotent; a check needs a change to look for |
FR-004 | check | bun run verify | suite/governance-action.test.ts: a destructive action is asked, or confirmed by the flag where a flag may, or refused; a preview never asks and never changes; suite/governance-action.test.ts: for any action, surface and invocation: nothing that is not a preview changes unless it was decided to proceed |
FR-005 | check | bun run verify | suite/governance-action.test.ts: for any action, surface and invocation: nothing that is not a preview changes unless it was decided to proceed; suite/governance-action.test.ts: a machine surface refuses an action for a person and an option only a person may give |
FR-006 | check | bun run verify | suite/governance-action.test.ts: the help says whether an action is safe to repeat: the fact is declared and defaults to the safe side |
FR-007 | check | bun run verify | suite/governance-action.test.ts: the flags make one mode, both together are refused, and only apply may change |
FR-008 | check | bun run verify | suite/governance-problems.test.ts: a problem has one of four levels, holds only the fields given, and the counts are of its levels; suite/governance-problems.test.ts: a stable code is E or P then upper-case words, and each level has a generic code that a program may replace |
FR-009 | check | bun run verify | suite/governance-problems.test.ts: the exit table: every status has one name, every name one status and a meaning; the ranges are the ones decided |
FR-010 | check | bun run verify | suite/governance-problems.test.ts: a run ends with findings where anything failed, pending only where asked, else ok; a warning never changes it |
FR-011 | check | bun run verify | suite/governance-problems.test.ts: redaction masks URL passwords, secret settings and given secrets, longest first, and leaves other text alone |
0003 Headless output
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/logical-model.test.ts: a program is a tree of branches and commands; each command has its path, its id and the facts of its action; suite/logical-model.test.ts: the defects of a tree are found: duplicate words, an action that contradicts itself, reserved options and shorts, an unknown human-only option, a built-in word; suite/logical-model.test.ts: what a command tells while it runs is a progress event, and the writer of a phase says each kind |
FR-002 | check | bun run verify | suite/logical-model.test.ts: an outcome is a value with empty defaults; merging keeps the first's data and text, orders the rest, and shows a hint once; suite/logical-model.test.ts: what a command tells while it runs is a progress event, and the writer of a phase says each kind |
FR-003 | check | bun run verify | suite/logical-model.test.ts: the flags choose a verbosity: json over all, quiet with verbose refused; a table says what each shows |
FR-004 | check | bun run verify | suite/logical-closing.test.ts: the closing block is in the order problems, your move, pending, hints, steps; the first failure is the verdict and is not listed twice; suite/logical-closing.test.ts: where nothing failed the verdict states counts and never 'all well'; the next command is the first step for the person, else the suggested one; suite/logical-closing.test.ts: pending is never a failure and shows its next step (the case of stolos 052); suite/logical-closing.test.ts: for any problems: the verdict and the counts agree, a failure is listed once, and there is at most one next command |
FR-005 | check | bun run verify | suite/logical-closing.test.ts: a fix is split into its one command and the words around it (the golden cases of stolos) |
FR-006 | check | bun run verify | suite/logical-told.test.ts: an envelope is one document with a fixed order of keys, whose status, counts and data agree with the run; suite/logical-told.test.ts: toJson never throws and gives what JSON can say, for any value |
FR-007 | check | bun run verify | suite/logical-told.test.ts: SARIF 2.1.0 holds the findings returned, else those of the failures and warnings, with rules, levels and locations |
FR-008 | check | bun run verify | suite/logical-told.test.ts: a raw error of another tool is given its code and fix by the first pattern that matches, and an unknown one is left as it is; suite/logical-told.test.ts: explain gives one explanation for a code in either case, the list without one, and names the nearest for one that is not there; suite/logical-told.test.ts: a catalogue is checked: stable codes, no blank field, patterns that name a code explained |
FR-009 | check | bun run verify | suite/logical-told.test.ts: a hint is due when never shown or shown seven days ago, once for each id, and the memory changes only where one is due |
FR-010 | check | bun run verify | suite/logical-told.test.ts: a nested run hands its summary to its parent, which reads it back by a decoder and merges it so that every problem is there once |
FR-011 | check | bun run verify | suite/logical-told.test.ts: every text of an outcome and of its progress passes through the redactor, however deep, and doing it twice is doing it once |
FR-012 | check | bun run verify | suite/logical-model.test.ts: every command takes the standard options by its effect, and the key of each is a long option |
0004 Headless command line
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/cli-render.test.ts: every tone is one mark of the palette, with its ASCII fallback where there are no emoji; suite/cli-render.test.ts: no file of the layers writes an emoji of its own: the only marks are the palette's; suite/cli-render.test.ts: the closing block is drawn with the palette's marks in the order it was chosen, the verdict last, and in quiet only the verdict |
FR-002 | check | bun run verify | suite/cli-render.test.ts: colour is written only where the look allows it: a terminal, no NO_COLOR, FORCE_COLOR, and not with --no-color; emoji by the locale and the program's own variables; suite/cli-render.test.ts: a line is cut to the terminal's width, never inside a character or an escape sequence |
FR-003 | check | bun run verify | suite/cli-render.test.ts: the closing block is drawn with the palette's marks in the order it was chosen, the verdict last, and in quiet only the verdict; suite/cli-render.test.ts: in calm a phase that went well is no line and one with something to see is one, with its honest mark; verbose shows every step and item; suite/cli-render.test.ts: a phase with nothing but a step for the person shows that mark, and one with a failure and a warning the failure's; suite/cli-run.test.ts: golden: an action with nothing to say ends with the verdict alone, on the error stream, with status 0; suite/cli-run.test.ts: golden: a warning and something pending are shown with their codes and one command each, and the run still ends with 0; suite/cli-run.test.ts: golden: a failure is the verdict once, with its code and its one command, and the status is 1; quiet is the verdict alone; suite/cli-run.test.ts: golden: --verbose shows every step as it happens, the green phase line, and the log's path |
FR-004 | check | bun run verify | suite/cli-run.test.ts: golden: an action with nothing to say ends with the verdict alone, on the error stream, with status 0; suite/cli-run.test.ts: golden: a warning and something pending are shown with their codes and one command each, and the run still ends with 0; suite/cli-run.test.ts: golden: a failure is the verdict once, with its code and its one command, and the status is 1; quiet is the verdict alone; suite/cli-run.test.ts: a report's text goes to the output stream and its narration to the error stream; a report with nothing to say beyond its text writes no block; suite/cli-run.test.ts: --json is one document and nothing for a person; --sarif the same as SARIF; neither with colour; suite/cli-run.test.ts: a secret the program holds never reaches the output, the error, the log or the JSON, however a command tells it |
FR-005 | check | bun run verify | suite/cli-render.test.ts: the spinner and the bar are pure functions of a number, with ASCII frames where there are no emoji; suite/cli-render.test.ts: on a terminal the phase in progress is one line redrawn in place and erased before a line that stays; elsewhere there are only the final lines; suite/otel.test.ts: the surface does not import the telemetry: the observer is a port, and only the adapter names the telemetry |
FR-006 | check | bun run verify | suite/cli-run.test.ts: a destructive action asks the person where there is one, runs with --yes, and is refused with no terminal and no --yes; suite/cli-run.test.ts: ask and confirm ask no one without a terminal, take one line with one, and only a plain yes is yes |
FR-007 | check | bun run verify | suite/cli-run.test.ts: a run writes its log in the folder of state, with every item and problem whatever the verbosity, and a problem carries the number of its line; suite/cli-state.test.ts: a log is named by its time and process, and every line of it has its time, its kind and its number; suite/cli-state.test.ts: logs older than 30 days are pruned, and nothing else; the newest log is found by its time; suite/cli-state.test.ts: the file system's store makes folders, appends, lists files with their time, removes, and never throws; the pruning works on a real folder |
FR-008 | check | bun run verify | suite/cli-run.test.ts: a hint is shown once in seven days from the memory in the folder of state, always with --verbose, never with --quiet, and every time where there is no state; suite/cli-state.test.ts: the hints' memory is read as nothing where it is missing or wrong, written as read, and due hints are written once they are shown |
FR-009 | check | bun run verify | suite/cli-process.test.ts: a real SIGINT ends a running program with 130 and a real SIGTERM with 143, and the program says Stopped; suite/cli-run.test.ts: SIGINT ends the run with 130 and SIGTERM with 143, aborts the signal of the command, ends with Stopped, and leaves no listener |
FR-010 | check | bun run verify | suite/cli-help.test.ts: the help of every level is made from the declaration, and a verb's help says its effect, its repetition and its audience, with the options of its effect; suite/cli-help.test.ts: the completion table has every command and every option of the tree, and its words are all words the parser accepts; suite/cli-help.test.ts: the completion scripts are made for bash, zsh, fish and PowerShell, hold every word of the table, and the bash script is valid bash; suite/cli-help.test.ts: the man page lists each command with its effect and options and the exit statuses of the one table, and holds no date; suite/cli-help.test.ts: help is shown by --help, -h and `help`, at every level, and the same text each way |
FR-011 | check | bun run verify | suite/cli-help.test.ts: help is shown by --help, -h and `help`, at every level, and the same text each way; suite/cli-process.test.ts: the host binds the world of the process and nothing else does: the environment, the streams' kind, the clock, the signals, the timer and a line of input; suite/cli-run.test.ts: golden: an action with nothing to say ends with the verdict alone, on the error stream, with status 0; suite/cli-run.test.ts: --json is one document and nothing for a person; --sarif the same as SARIF; neither with colour; suite/cli-run.test.ts: the surface answers --version, help, explain, completions and man itself, and runs the default command where no word is given; suite/cli-run.test.ts: a run touches no process global: the same world gives the same result, whatever the process's own environment is; suite/cli-run.test.ts: a destructive action asks the person where there is one, runs with --yes, and is refused with no terminal and no --yes; suite/cli-run.test.ts: a nested run writes its summary to its parent's file and draws nothing; the parent reads it back; suite/cli-run.test.ts: a secret the program holds never reaches the output, the error, the log or the JSON, however a command tells it |
FR-012 | check | bun run verify | suite/cli-run.test.ts: a failure before a command runs is one sentence on the error stream and status 2, in every verbosity; a command that throws is one failure; suite/cli-run.test.ts: a nested run writes its summary to its parent's file and draws nothing; the parent reads it back |
FR-013 | check | bun run verify | suite/cli-process.test.ts: a failure to start is typed, has the status of its kind in the range 64 to 78, and is one line and a fix line; suite/cli-process.test.ts: a launcher that cannot start gives one line and the status of its kind; one that can runs the program with the process's world; suite/cli-process.test.ts: a check of the start that throws is a failure of the kind software, and a launch can be given its own arguments; suite/cli-process.test.ts: the host binds the world of the process and nothing else does: the environment, the streams' kind, the clock, the signals, the timer and a line of input |
FR-014 | check | bun run verify | suite/cli-run.test.ts: a run writes its log in the folder of state, with every item and problem whatever the verbosity, and a problem carries the number of its line; suite/cli-state.test.ts: the folder of state is the system's: XDG on Linux and macOS, the known folders on Windows, and none where there is no home |
FR-015 | check | bun run verify | suite/otel.test.ts: a run is a headless.run span with the program, the command, the status and the counts, a phase event for each phase and a problem log for each problem; suite/otel.test.ts: a run whose status is not 0 ends its span as an error, with the status named; a refusal before the command is no run; suite/otel.test.ts: the surface does not import the telemetry: the observer is a port, and only the adapter names the telemetry |
0005 Headless reach and targets
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/reach-targets.test.ts: a reach policy has the safe defaults, and the decision refuses a channel it does not list and a changing action for a key that may only read |
FR-002 | check | bun run verify | suite/reach-targets.test.ts: the classes of target are four, the trust policy has one value for each field and names no secret |
FR-003 | check | bun run verify | suite/reach-targets.test.ts: an action with no reach policy is rejected, and a policy that contradicts its action is found by rule |
FR-004 | check | bun run verify | suite/reach-targets.test.ts: an executor has the shape of the system's proc: proc.output is an Executor's exec, and the local target is it |
FR-005 | check | bun run verify | suite/reach-targets.test.ts: the remote command is quoted word by word into one string that a real shell reads back as the same words; suite/reach-targets.test.ts: the ssh arguments hold the one set of options in one place, put the destination after --, and never accept a host key; suite/reach-targets.test.ts: an address that could be taken for an option, a user, a port, a jump or a credential that is not one is not a target; suite/reach-targets.test.ts: the ssh executor runs the system's client with the quoted command, gives a call that cannot be made as 255 and runs a script as sh -s with it on the input |
FR-006 | check | bun run verify | suite/reach-targets.test.ts: the ssh arguments hold the one set of options in one place, put the destination after --, and never accept a host key; suite/reach-targets.test.ts: a host is learned only by the fingerprint a person checked, and only if the host showed it |
FR-007 | check | bun run verify | suite/reach-targets.test.ts: --on dispatches the same verb as run --json to each target; one that cannot be reached is a row, the report is one, and the status is 1; suite/reach-targets.test.ts: a verb reaches targets only where its reach policy allows ssh, and the selector, the executors and the forwarding are checked; suite/reach-targets.test.ts: a fan-out is bounded by a cap and a time: at most the cap run at once, a host that hangs is a time out, one that throws is a row, and every id has its row in order; suite/reach-targets.test.ts: the entry a target runs: run --ssh takes the command a forced command was given, refuses all else, a verb with no reach or no ssh channel, a changing verb for a read-only key, and any onward hop |
FR-008 | check | bun run verify | suite/reach-targets.test.ts: the entry a target runs: run --ssh takes the command a forced command was given, refuses all else, a verb with no reach or no ssh channel, a changing verb for a read-only key, and any onward hop; suite/reach-targets.test.ts: the inbound lines are generated from reach: forced commands with restrict, read-only for a read-only key, no line for a mode nothing is reached through, and a Match block that takes away what a forced command does not need |
FR-009 | check | bun run verify | suite/reach-targets.test.ts: --on dispatches the same verb as run --json to each target; one that cannot be reached is a row, the report is one, and the status is 1; suite/reach-targets.test.ts: an exit of 255, a host key that does not match, a time out, a program that is not there, a refusal and a body that is not a report each have their code, and the statuses of the runs over several are 0, 3, 2 or 1; suite/reach-targets.test.ts: the runs over several targets are one report: the table of the hosts, each host's data under its id, every problem once and marked with its target |
FR-010 | check | bun run verify | suite/reach-targets.test.ts: a hand-built ssh argument in a consumer is found, and one made by the executor is not asked of the consumer; suite/reach-targets.test.ts: no unit of the library imports an ssh library or reads a protocol: the rows of the layers table for targets and for the inbound surface allow the terminal, the system and Core only |
FR-011 | check | bun run verify | suite/container-target.test.ts |
0006 Headless machine verbs
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/machine-verbs.test.ts: the branch machine has query and check, each a read-only report reached through every channel, and the model has no defect with the reach policy required; suite/machine-verbs.test.ts: each verb is dispatched to the targets of a selector as run --json machine <verb> |
FR-002 | check | bun run verify | suite/machine-verbs.test.ts: a query is handed to the backend byte for byte and never read; a wrong number of words is a failure with its code and the backend is not called; suite/machine-verbs.test.ts: a check id is one line with no NUL, an id that is not one of the checks the backend names is refused with the nearest, and a failing check is the backend's outcome |
FR-003 | check | bun run verify | suite/machine-verbs.test.ts: no unit of the library names data, grc or the ontology: the layers check finds such an import, and the real tree has none |
0007 Headless ssh coverage
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/ssh-coverage.test.ts: the three families hold what each is for; a terminal and the interactive family need the reach policy to say so |
FR-002 | check | bun run verify | suite/ssh-coverage.test.ts: trust-host shows what a host presents, ed25519 first, records nothing without a fingerprint or with --dry-run, and records only a fingerprint the host showed; suite/ssh-coverage.test.ts: the scan gives fingerprints with ed25519 first, and a fingerprint that was not shown writes nothing |
FR-003 | check | bun run verify | suite/ssh-coverage.test.ts: scp copies with the options of the family, -P, --, every word its own argument, a local path that could be a host made relative, an IPv6 address in brackets, and a remote path a shell could expand refused; suite/ssh-coverage.test.ts: the local executor copies files with the system's file functions and gives a failure as an exit |
FR-004 | check | bun run verify | suite/ssh-coverage.test.ts: a session is a spawned client with piped streams: the quoted command reaches it, lines written come back, and it ends with its status when the input closes |
FR-005 | check | bun run verify | suite/ssh-coverage.test.ts: the scan gives fingerprints with ed25519 first, and a fingerprint that was not shown writes nothing; suite/ssh-coverage.test.ts: the key tools have typed argument vectors, and what they write is parsed: keys deduplicated with ed25519 first, fingerprints with size, comment and type |
FR-006 | check | bun run verify | suite/ssh-coverage.test.ts: the known hosts are two files in one option, the repository's first and the user's after, one file in the pinned family, and a path with a space is refused |
FR-007 | check | bun run verify | suite/ssh-coverage.test.ts: every way stolos calls ssh, in a table, is built, with the options its shape needs and never accept-new |
0008 Headless exit, style and consent
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/exit-style-consent.test.ts: a command names its exit and the run ends with it; the status and the envelope agree, and a failure is never hidden by a name of ok or pending |
FR-002 | check | bun run verify | suite/exit-style-consent.test.ts: the marks are plain unless the stream is a terminal that shows emoji and nothing asks for plain; Io.marks decides explicitly |
FR-003 | check | bun run verify | suite/exit-style-consent.test.ts: the redactor replaces secrets and tokens of a known shape and declared keys, and leaves a sentence that names a variable alone; suite/redact-references.test.ts: a value that a shell or a script expands is left and a literal is replaced |
FR-004 | check | bun run verify | suite/exit-style-consent.test.ts: an action of the class consent goes on with no terminal on a consent given before, asks a person where there is one, is refused otherwise, and records the consent in the log |
0009 Headless composition
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/composition.test.ts: a plugin is a value made by definePlugin with a stable id, and every problem of a definition is returned as a value |
FR-002 | check | bun run verify | suite/composition.test.ts: a plugin is a value made by definePlugin with a stable id, and every problem of a definition is returned as a value |
FR-003 | check | bun run verify | suite/composition.test.ts: composing a host and plugins gives one program with the owner of every command, the same in any order of the plugins, with one help, completions, man page and verb ids |
FR-004 | check | bun run verify | suite/composition.test.ts: two plugins claiming one verb is a conflict value naming both owners, no winner is chosen, and the other errors are values too |
FR-005 | check | bun run verify | suite/composition.test.ts: the composed program is checked as one: a command with no reach policy and a tier that is not on the ladder are errors of the composition |
FR-006 | check | bun run verify | suite/composition.test.ts: composing a host and plugins gives one program with the owner of every command, the same in any order of the plugins, with one help, completions, man page and verb ids; suite/composition.test.ts: a composed program runs as one: its commands run, the run log names the plugin they came from, and there is one log |
FR-007 | check | bun run verify | suite/composition.test.ts: the conformance suite gives a result for every rule, passes a good plugin and fails, by rule, a plugin that breaks one |
0010 Headless audience tiers
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/audience-tiers.test.ts: an action has the lowest tier by default, a tier name is parsed, and a ladder is data with every problem returned as a value |
FR-002 | check | bun run verify | suite/audience-tiers.test.ts: the gate is pure and total: a caller sees what is at or below its tier, and a tier not on the ladder hides the action and never exposes it |
FR-003 | check | bun run verify | suite/audience-tiers.test.ts: the tree a caller sees has no hidden verb and no branch left empty, and the help, the completions and the man page are drawn from it |
FR-004 | check | bun run verify | suite/audience-tiers.test.ts: invoking a hidden verb is refused with the typed reason and runs nothing, the help of the command line shows only what the tier may see, and an unknown word stays unknown |
FR-005 | check | bun run verify | suite/audience-tiers.test.ts: the tier does not change what the effect needs: a destructive action of a high tier is still confirmed, and a program for a person is still refused by a machine surface |
FR-006 | check | bun run verify | suite/audience-tiers.test.ts: invoking a hidden verb is refused with the typed reason and runs nothing, the help of the command line shows only what the tier may see, and an unknown word stays unknown |
0011 Headless credential prompt
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/credential-prompt.test.ts: a secret is never written: its text, JSON and inspected forms are redacted and only reveal() gives it, and an id is parsed |
FR-002 | check | bun run verify | suite/credential-prompt.test.ts: a credential is asked once for each command, shared by asks at the same time, in turn for different ids, and dropped at close |
FR-003 | check | bun run verify | suite/credential-prompt.test.ts: where no person can be asked the refusal is a typed value and the port is not called, and every refusal has a sentence and a code with no value in it |
FR-004 | check | bun run verify | suite/credential-prompt.test.ts: a prompt has a hard time limit that aborts the port, a stopped run cancels it, and an empty or short answer is refused |
FR-005 | check | bun run verify | suite/credential-prompt.test.ts: a command is given its session, the redaction of the run covers the values it handed out, the log records the prompt and never the value, and the session ends with the command |
FR-006 | check | bun run verify | suite/credential-prompt.test.ts: a command is given its session, the redaction of the run covers the values it handed out, the log records the prompt and never the value, and the session ends with the command |
FR-007 | check | bun run verify | suite/credential-prompt.test.ts: the terminal adapter draws the prompt, reads with no echo for a secret and with echo for a yes, never writes the value, and cancels on an ended input; the reader is pure and the devices are rows by system |
FR-008 | check | bun run verify | suite/credential-prompt.test.ts: the helper ssh runs reads the request from its argument, asks once, writes the answer as one line and ends 0, ends 1 with only a reason where there is none, and ends 2 for arguments that are no request |
FR-009 | check | bun run verify | suite/credential-prompt.test.ts: the launcher is a script on Linux and macOS and a .cmd file on Windows, a path that cannot be quoted is refused, and ssh is told to use it even with a terminal |
0012 Headless executor check mode
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/executor-check-mode.test.ts: a plan is made only from valid steps and every problem is returned as a value |
FR-002 | check | bun run verify | suite/executor-check-mode.test.ts: check runs only check vectors and reports a step with none as unsupported, apply runs apply vectors, a failure skips the rest, and a throwing executor fails a step and not the run |
FR-003 | check | bun run verify | suite/executor-check-mode.test.ts: check runs only check vectors and reports a step with none as unsupported, apply runs apply vectors, a failure skips the rest, and a throwing executor fails a step and not the run |
FR-004 | check | bun run verify | suite/executor-check-mode.test.ts: the canonical form is stable and the hash is SHA-256 of it: the published vectors hold, any change to the plan changes the hash, and the run options are not in it |
FR-005 | check | bun run verify | suite/executor-check-mode.test.ts: an apply names its approval: a hash that is not the plan's is refused with nothing run, the right hash and an explicit unsigned go on |
FR-006 | check | bun run verify | suite/executor-check-mode.test.ts: a run streams started, line, finished and progress in order, from an executor that streams and from one that does not, and a surface can render the plan and the events |
FR-007 | check | bun run verify | suite/executor-check-mode.test.ts: a run streams started, line, finished and progress in order, from an executor that streams and from one that does not, and a surface can render the plan and the events |
0013 Headless tool server (MCP)
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/mcp-rest.test.ts: the tools are generated from governance: one for each visible command that may come through mcp and is not for a person alone, with a schema from its options and hints from its effect |
FR-002 | check | bun run verify | suite/mcp-rest.test.ts: a call is gated as a machine's: a hidden tool is refused with the reason of the tier, a channel, a key that may only read, what only a person may give or do and a destructive action that a flag cannot confirm are refused with theirs, and a preview goes |
FR-003 | check | bun run verify | suite/mcp-rest.test.ts: a call is gated as a machine's: a hidden tool is refused with the reason of the tier, a channel, a key that may only read, what only a person may give or do and a destructive action that a flag cannot confirm are refused with theirs, and a preview goes |
FR-004 | check | bun run verify | suite/mcp-rest.test.ts: the server answers initialize, ping and tools/list, never answers a notification, and gives the codes of JSON-RPC for a line that is not JSON, not a request, a method it does not serve and parameters that are wrong |
FR-005 | check | bun run verify | suite/mcp-rest.test.ts: the arguments of a call become a command line: flags, values as --name=value, operands last, and an unknown, mistyped, missing or option-like argument is an error value; suite/mcp-rest.test.ts: a call runs the command through the command line with the caller's tier and gives its JSON document as the text and as the structured content |
FR-006 | check | bun run verify | suite/mcp-rest.test.ts: the server answers initialize, ping and tools/list, never answers a notification, and gives the codes of JSON-RPC for a line that is not JSON, not a request, a method it does not serve and parameters that are wrong |
0014 Headless REST surface (skeleton)
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/mcp-rest.test.ts: the OpenAPI 3.1 document has a path for each command the tier may see and the rest channel allows, GET for a report and POST for the rest, with the effect and the tier of each, and the refusals as data |
FR-002 | check | bun run verify | suite/mcp-rest.test.ts: the OpenAPI 3.1 document has a path for each command the tier may see and the rest channel allows, GET for a report and POST for the rest, with the effect and the tier of each, and the refusals as data |
FR-003 | check | bun run verify | suite/mcp-rest.test.ts: the OpenAPI 3.1 document has a path for each command the tier may see and the rest channel allows, GET for a report and POST for the rest, with the effect and the tier of each, and the refusals as data |
0015 Headless describes a program
| Requirement | Mechanism | Who | Note |
|---|---|---|---|
FR-001 | check | bun run verify | suite/describe.test.ts: describe gives one model for each tier of the ladder, lowest first, with the three layers, pure and deterministic, and describeChecked says why where the name is not an id; suite/describe.test.ts: for any generated program, each model has unique ids, a trace between its elements, the tools and routes of the surfaces for its tier, a command for each visible verb and nothing dropped |
FR-002 | check | bun run verify | suite/describe.test.ts: every element has the layer of its kind and an id made by the agreed forms: verbs, commands, tools, routes, options and operands |
FR-003 | check | bun run verify | suite/describe.test.ts: every element has the layer of its kind and an id made by the agreed forms: verbs, commands, tools, routes, options and operands |
FR-004 | check | bun run verify | suite/describe.test.ts: governance holds the action of each verb with its annotations, the problem codes with their explanation and the exit statuses by name |
FR-005 | check | bun run verify | suite/describe.test.ts: the logical layer holds the program, the verbs with their options, standard ones included, operands and examples, and the plugin that contributed a verb as its parent |
FR-006 | check | bun run verify | suite/describe.test.ts: the problems of the model are findings of the element they are about, and nothing is thrown; suite/describe.test.ts: two verbs that would be one tool are one tool and a finding, never two elements of one id |
FR-007 | check | bun run verify | suite/describe.test.ts: the tools of a model are those the tool server serves for the tier and the routes are the paths of the OpenAPI document, drawn by the reach and the audience; suite/describe.test.ts: for any generated program, each model has unique ids, a trace between its elements, the tools and routes of the surfaces for its tier, a command for each visible verb and nothing dropped |
FR-008 | check | bun run verify | suite/describe.test.ts: the trace goes from the action to the verb to each projection, every visible verb traces to a command, and a verb a tier hides is there with no projection and says why; suite/describe.test.ts: for any generated program, each model has unique ids, a trace between its elements, the tools and routes of the surfaces for its tier, a command for each visible verb and nothing dropped |
FR-009 | check | bun run verify | suite/describe.test.ts: the trace goes from the action to the verb to each projection, every visible verb traces to a command, and a verb a tier hides is there with no projection and says why |
FR-010 | check | bun run verify | suite/describe.test.ts: every annotation of every action, problem code and exit status ends in a disposition, so that the check of Kanon finds nothing dropped; suite/describe.test.ts: for any generated program, each model has unique ids, a trace between its elements, the tools and routes of the surfaces for its tier, a command for each visible verb and nothing dropped |
FR-011 | check | bun run verify | suite/describe.test.ts: where an element is written is what the caller says, and the name of the program or of the plugin and the line 1 where it says nothing |