Module cli 

CLI grammar: launch syntax, wrapper options and reserved command words (spec §5; SP1 design §4) and the repository commands (SP3 design §7).

Modules

completions 🔒

The shell completion model. The model exists only for clap_complete::generate: it never parses an argument line, and it mirrors the grammar table on CommandWord (scope, flags, positionals), so the generated scripts cannot drift from the hand dispatcher.

Structs

Cli 🔒

Select and launch profiles for multiple coding agents.

Enums

AgentScope 🔒

Where a command word may appear: with an agent, without one, or both.

Command 🔒

Every reserved word — the eight of spec §5.3 plus resolve, current, repositories, doctor and completions — is dispatched before Clap (SP3 design §7.2), so Clap sees only the agent form.

CommandWord 🔒

The repository commands (SP3 design §7.1).

Invocation 🔒

The outcome of the argument checks (SP1 design §4.2, SP3 design §7.2).

LaunchDiscovery 🔒

Whether a launch runs discovery (SP3 design §7.6).

Positional 🔒

The positional words a command takes after its command word.

Token 🔒

One token before the first --, after --repo binding (SP3 design §7.2 step 1).

Constants

COMMANDS_HELP 🔒

LAUNCH_USAGE 🔒

REPO_EQUALS 🔒

Functions

bare_words 🔒

base_dir 🔒

The directory a --repo value is joined to. An absolute --repo needs no current directory, so a command given one still works from a working directory that no longer exists (design §6.2, §10 “--repo is the explicit override”).

bind 🔒

Step 1: --repo=<v> carries its value; a bare --repo consumes the next token whatever it is.

confirm 🔒

Asks the user a yes/no question: prints question to stderr without a newline, flushes, and reads one line. Only y or yes, trimmed and ignoring ASCII case, counts; EOF or anything else is no.

current_dir 🔒

cut 🔒

Splits at the first --: the tokens before it, and the opaque arguments when there is one.

execute 🔒

has_help 🔒

has_version 🔒

launch_discovery 🔒

no_profile 🔒

option_name 🔒

Only the part before = is echoed: the value may be a secret meant for the agent (spec §36).

options 🔒

parse_profile 🔒

repo_value 🔒

The value of a --repo=<v> token, which may be non-UTF-8.

report 🔒

run

Runs the CLI and returns the process exit code. This is the only place an exit code is decided.

run_agents 🔒

agents (SP5a design §3): every registry entry with its executable evidence. Read-only — the load in run_command refuses an invalid configuration but nothing is created here (V3 §17) — and it always exits 0: a missing or unusable executable is listed as a fact, not an error, while any discovery error that is not one of those facts propagates.

run_command 🔒

SP3 design §7.4 steps 2-5.

run_create 🔒

<agent> create <profile> (SP5a design §3): materializes the adapter’s paths exactly once. Both outcomes exit 0; the configuration was already refused if invalid by run_command’s load, and is otherwise unused.

run_delete 🔒

<agent> delete <profile> [--yes] (SP5a design §3). An advisory referenced-profile check first, then --yes or an interactive confirmation, then the locked lifecycle::delete. A missing profile exits 0 rather than erroring, because that is the state the user asked for — the same line and code as a profile that vanishes between the check and the lock. The configuration was already refused if invalid by run_command’s load, and is read again under the lock inside lifecycle::delete.

run_doctor 🔒

SP5c design §3-§4: every check runs and prints; the exit code is 5 when any record failed. A working directory that cannot be read is not an error here: discovering . then yields the repository record.

run_launch 🔒

run_profiles 🔒

profiles and bare list and <agent> list (SP5a design §3): the read-only listings. Nothing is created — run_command has already loaded the configuration (V3 §17) but no root, no config.toml and no lock is written here — and every outcome exits 0: an empty listing is the answer, not an error, and unrecognized on-disk entries are a stderr note rather than a failure.

run_repositories 🔒

repositories (SP5b design §4): every stored mapping with its classification. Read-only, takes no lock, and always exits 0: an entry that cannot be examined is reported, not raised.

run_unlink 🔒

unlink chooses its keys without discovery when --repo is given (SP3 design §6.2).

split 🔒

agent-profile <agent> …: SP1 design §4.2 rules 3-4 as replaced by SP3 design §7.2.

split_top_level 🔒

agent-profile <command> … for the reserved words dispatched before Clap (SP3 design §7.2).

usage 🔒

validate_command 🔒

Step 5: bare holds the bare words after the command word.

with_unknown_configured 🔒

Fills unknown_configured on UnknownAgent and AgentNotInstalled. With no configuration at hand, one best-effort load is attempted; its failure leaves the list empty (design §4.2 check 3).

write_lines 🔒

write_out 🔒