Documentation
¶
Index ¶
- Variables
- func Command(args []string, in io.Reader, out, errOut io.Writer) int
- func MenuCommand(args []string, in io.Reader, out, errOut io.Writer, known []primitive.Known) int
- func MigrateCommand(out, errOut io.Writer) int
- type Call
- type Draft
- type Reader
- type Report
- type ReviewActions
- type ReviewConfig
- type Session
Constants ¶
This section is empty.
Variables ¶
var ( Run = pipeline.Run DefaultOptions = pipeline.DefaultOptions DefaultReaders = history.DefaultReaders Review = routine.Review WriteFunnel = routine.WriteFunnel WriteText = routine.WriteText )
var Version = ""
Version is the runner's version, shown beside the logo. The runner sets it.
Functions ¶
func Command ¶
Command is `tap discover`: read this machine's agent session history, report how it narrowed to primitives, and optionally pick primitives to save. It needs no account and sends nothing anywhere. args exclude the command name. It returns the process exit code.
func MenuCommand ¶
MenuCommand reads this machine's history, condenses it into primitives, composes them (with known, primitives fetched by a caller such as the Telara CLI, plus those accepted here before) and shows the menu. --all accepts every proposed primitive without asking.
func MigrateCommand ¶
MigrateCommand is `tap discover migrate-saved`: primitives saved as full packages in agents' skills folders (before the TAP collection) move into the collection and leave a pointer behind.
Types ¶
type ReviewActions ¶
type ReviewActions = routine.ReviewActions
type ReviewConfig ¶
type ReviewConfig = routine.ReviewConfig
Source Files
¶
Directories
¶
| Path | Synopsis |
|---|---|
|
Package author is the author path: Discover proposes, an agent writes the primitive, and the result is validated through the real runner before it is saved.
|
Package author is the author path: Discover proposes, an agent writes the primitive, and the result is validated through the real runner before it is saved. |
|
Package client is the one list of coding agents discover knows about: their names, how to tell one is installed, where their skills folders are, how the TAP MCP server is connected to them, and whether the runner can borrow their connections.
|
Package client is the one list of coding agents discover knows about: their names, how to tell one is installed, where their skills folders are, how the TAP MCP server is connected to them, and whether the runner can borrow their connections. |
|
cmd
|
|
|
discover
command
Command discover runs `tap discover` from this module: it reads this machine's agent history and ends in the primitive menu.
|
Command discover runs `tap discover` from this module: it reads this machine's agent history and ends in the primitive menu. |
|
discover-eval
command
Command discover-eval freezes a corpus of local agent history, runs discover on it, and draws the labeling sample.
|
Command discover-eval freezes a corpus of local agent history, runs discover on it, and draws the labeling sample. |
|
discover-fixture
command
Command discover-fixture captures a reader test fixture from a real session on this machine.
|
Command discover-fixture captures a reader test fixture from a real session on this machine. |
|
Package codegen turns recurring logic into an exact, reviewable TAP program without calling a model: the program graph, its variants, and the generated package.
|
Package codegen turns recurring logic into an exact, reviewable TAP program without calling a model: the program graph, its variants, and the generated package. |
|
Package eval is the evaluation corpus: reproducible samples of task episodes from a frozen corpus, labeled from their own evidence, and the request-level episode selection.
|
Package eval is the evaluation corpus: reproducible samples of task episodes from a frozen corpus, labeled from their own evidence, and the request-level episode selection. |
|
Package genreview shows a generated program in full and takes one local decision: accept, deny or refine it, and runs the model-free generate command.
|
Package genreview shows a generated program in full and takes one local decision: accept, deny or refine it, and runs the model-free generate command. |
|
Package history reads the session history agent clients keep on this machine: Claude Code transcripts, Codex rollouts, Cursor's chat store, and frozen corpora.
|
Package history reads the session history agent clients keep on this machine: Claude Code transcripts, Codex rollouts, Cursor's chat store, and frozen corpora. |
|
internal
|
|
|
testkit
Package testkit holds the corpus builders and fakes the discover packages' tests share.
|
Package testkit holds the corpus builders and fakes the discover packages' tests share. |
|
Package model holds the data types a discover run produces: the report, its candidates, routines, contracts, drafts and span groups, with their enums.
|
Package model holds the data types a discover run produces: the report, its candidates, routines, contracts, drafts and span groups, with their enums. |
|
Package pack writes a draft as a TAP package and installs it the way an agent client finds a skill.
|
Package pack writes a draft as a TAP package and installs it the way an agent client finds a skill. |
|
Package pipeline runs a discover pass end to end: read the history, mine and score patterns, group them into candidates, and assemble the report.
|
Package pipeline runs a discover pass end to end: read the history, mine and score patterns, group them into candidates, and assemble the report. |
|
Package primitive condenses agent history into proposed primitives.
|
Package primitive condenses agent history into proposed primitives. |
|
Package pyparse reads recorded inline Python as data.
|
Package pyparse reads recorded inline Python as data. |
|
Package redact keeps credentials in recorded session history out of drafts.
|
Package redact keeps credentials in recorded session history out of drafts. |
|
Package retrieval selects, groups and screens causal slices of recorded calls as candidates for human review.
|
Package retrieval selects, groups and screens causal slices of recorded calls as candidates for human review. |
|
Package routine finds recurring work in normalized sessions and decides what it is: mining, significance, request-level routines, their task contracts and states, drafting a TAP package, and the review of what passed.
|
Package routine finds recurring work in normalized sessions and decides what it is: mining, significance, request-level routines, their task contracts and states, drafting a TAP package, and the review of what passed. |
|
Package shellparse splits recorded shell command lines into simple commands and words.
|
Package shellparse splits recorded shell command lines into simple commands and words. |
|
Package termart draws the TAP logo, progress bars and live status rows that the CLI shows on a terminal.
|
Package termart draws the TAP logo, progress bars and live status rows that the CLI shows on a terminal. |
|
Package trace is the shared vocabulary of recorded agent sessions: calls, sessions, normalized steps and slots, and the step-effect helpers every later stage uses.
|
Package trace is the shared vocabulary of recorded agent sessions: calls, sessions, normalized steps and slots, and the step-effect helpers every later stage uses. |
|
Package util holds the small dependency-free helpers the discover packages share.
|
Package util holds the small dependency-free helpers the discover packages share. |