Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Temicide/thcode/llms.txt

Use this file to discover all available pages before exploring further.

The inspection commands give you a precise, read-only view of every meaningful dimension of a running thcode session. None of them mutate state, execute shell commands, or consume authority. Output is identical across interactive, redirected, linearized, and headless JSON surfaces — the same fields, the same canonical tokens, the same order.

/status — canonical session status

Alias: /st
Requires approval: No
/status renders the StatusProjection — the single serializable object that aggregates session health into one consistent snapshot.
FieldTypeDescription
workMode'plan' | 'build'Current Work Mode
permissionProfile'manual' | 'assisted' | 'full-access'Active permission profile
fullAccessbooleanWhether Full Access is currently enabled
providerIdstringActive provider identity
modelIdstringActive model identity
healthState'unconfigured' | 'configured' | 'checking' | 'available' | 'unavailable' | 'unhealthy' | 'quarantined'Provider health
contextPercentnumber | 'percentage unavailable'Context window utilization as a percentage, or the canonical token when no budget exists
enforcementVerifiedbooleanWhether authority enforcement is verified
The full ConversationProjection extends StatusProjection with the session identity (sessionId, name, workspaceRoot, lastActivity), the transcript, and the context projection. /status renders the status slice specifically. contextPercent uses the canonical measurement token 'percentage unavailable' when no real budget is configured — it is never shown as 0% or omitted.

/permissions — versioned PermissionMatrix

Alias: /perm
Requires approval: No
/permissions displays the AuthorityProjection — the current state of every authority dimension for this activation.
FieldTypeDescription
activationIdstringRuntime activation identity
activationRevisionnumberMonotonic revision counter for the activation
workspaceIdstringWorkspace binding
workMode'plan' | 'build'Current Work Mode
permissionProfile'manual' | 'assisted' | 'full-access'Active permission profile
fullAccessbooleanWhether Full Access is enabled
sensitiveTransferOverridebooleanFull-Access-only sensitive-transfer override (separate authority from fullAccess)
activeBoundaryExpansionCountnumberCount of active (non-revoked, non-expired) Boundary Expansions
enforcementVerifiedbooleanWhether enforcement is verified
Each of these fields is independently addressable — none of them implies any other. workMode and permissionProfile are separate authority axes; fullAccess is a distinct flag from permissionProfile: 'full-access'.

/connections — provider and service health

Alias: /conn
Requires approval: No
/connections shows the health state and connection identity for every configured provider and service. The output uses the canonical healthState token set:
StateMeaning
unconfiguredProvider has no credentials or configuration
configuredCredentials present but health not yet verified
checkingHealth probe in progress
availableProvider is reachable and responding
unavailableProvider is unreachable
unhealthyProvider is reachable but returning errors
quarantinedProvider has been isolated due to repeated failures
Connection identity includes the provider ID and the workspace-scoped binding. No credentials or secrets are rendered.

/models — Typhoon model inspection

Requires approval: No
No aliases
/models is inspection-only for the Typhoon model set. It lists configured and available models with their current resolution state. Unresolved model pins are honestly labeled — they are never shown as available or working when they cannot be resolved. Entries that are catalogued but not yet available are not made invokable.
/models does not implement or invoke any Epic 4 Capability Registry or Specialist service. It reflects the current state of model resolution honestly, including gaps.

/tools — capability catalog and diagnostics

Requires approval: No
No aliases
/tools exposes the full capability registry: the catalog of built-in tools, their support levels, and their diagnostic state. It supports several sub-actions:

browse

List all registered capabilities with their current state.

search

Filter the catalog by name, category, or modality.

inspect

Show detailed state for a single capability entry.

enable / disable

Enable or disable a capability (registry control, not an Epic 4 invocation).

diagnose

Run a diagnostic probe on a capability.

retest

Re-run a previous diagnostic and update the state.
Each capability entry reports a CapabilityProjection:
interface CapabilityProjection {
  readonly id: string;
  readonly state: 'working' | 'Catalogued — Not available yet' | 'disabled' | 'unconfigured' | 'unhealthy' | 'quarantined';
  readonly category: string;
  readonly modality: string;
}
Entries with state 'Catalogued — Not available yet' are displayed but are never made invokable. The catalog is the source of truth; model-supplied capabilities, URLs, or installers are never registry authority.

/activity — grouped activity history

Alias: /act
Requires approval: No
/activity builds the ActivityProjection from the durable event journal and displays it grouped by Prompt Round and operation class.
FieldTypeDescription
sequencenumberDurable sequence number
eventIdstringStable event identity
promptRoundIdstring | nullPrompt Round that produced this record
operationIdstring | nullOperation that produced this record
operationClassstringClass: prompt, effect, authority, boundary, evidence, remote, health, capability, context
statusstringCanonical UX state token
safeSummarystringSecret-free plain summary
authorityRevisionnumber | nullAuthority revision at the time of this record
evidenceRefstring | nullEvidence reference identity
timestampstringISO-8601 timestamp
completeness'complete' | 'corrupt' | 'unavailable' | 'not-authoritative'Record completeness
Activity includes every proposed or executed call — including auto-permitted list, read, and search operations. The journal replay deduplicates by eventId so no call or terminal outcome appears twice. Corrupt or incomplete records are labeled with their completeness state and a recovery action; the history never claims to be complete when it is not. The ActivityProjection also carries:
  • truncatedCount — number of older records beyond the display cap
  • corruptCount — number of corrupt or unavailable records
  • completetrue only when both counts are zero

/context — Active Model Context inspection

Alias: /ctx
Requires approval: No
/context inspects the bounded Active Model Context — what is currently included in the model’s context window, how much capacity is used, and what has been omitted.
FieldTypeDescription
estimatedTokensnumberEstimated total tokens in the current context
effectiveCapacitynumber | 'percentage unavailable'Effective context capacity after reserves
utilizationPercentnumber | 'percentage unavailable'Utilization as a percentage, or the canonical unavailable token
pinnedTurnCountnumberNumber of pinned turns
contributorsContextContributorProjection[]Per-item contributor list (optional, may be omitted)
omissions{ itemId, reason }[]Items omitted from context with reason
manifestIdstring | nullContext manifest identity
measurementstringMeasurement quality token
Each contributor entry includes its role, inclusion mode, provenance, trust classification, estimated token count, and measured byte count. Items that were omitted due to capacity, trust policy, or duplication appear in the omissions list with their reason. The utilization band thresholds are:
BandThreshold
greenBelow 70%
amber70–84%
orange85–94%
red95%+
percentage unavailableNo real budget configured

/usage — cumulative provider usage

Requires approval: No
No aliases
/usage renders the UsageProjection — the cumulative token and call counts across all provider interactions in the current session.
FieldTypeDescription
inputTokensnumberTotal input tokens consumed
outputTokensnumberTotal output tokens produced
cachedInputTokensnumberSubset of input tokens served from cache
callsnumberTotal provider call count
measurementstringMeasurement source: provider-reported, locally-measured, estimated, fallback, unknown, or mixed
observationsUsageObservation[]Per-operation observations with operationId, modelId, source, and observedAt
Each observation entry carries operationId, modelId, source, and observedAt. When observations come from multiple sources, measurement is 'mixed'.

/check — dependency preflight

Requires approval: No
No aliases
/check re-runs the non-mutating dependency preflight. It probes the environment for required runtimes, compilers, package managers, and documented commands without installing or modifying anything. Each probe returns a PrerequisiteProbeResult:
StatusMeaning
verifiedPrerequisite found and version confirmed
prerequisite-blockerRequired dependency missing; platform-specific guidance shown
probe-failedCould not run the probe; result is inconclusive
incompatibleDependency found but version is incompatible
The overall preflight result is one of all-verified, blocked, or unknown. Guidance for blockers comes only from the maintained probe registry or verified project documentation — no model-supplied URLs or installers are ever used.
/check produces a fresh evidenceId per run. Evidence identities are never reused across runs, so each preflight result is independently attributable.

/mode — switch Work Mode

Requires approval: Yes
No aliases
/mode switches between plan and build Work Mode. Because this mutates authority, it may require interactive approval. Without a TTY, it fails closed and returns noTtyCommandBlocked() — no authority is held in suspension.
/mode plan
/mode build
Work Mode is an independently addressable authority field. Switching it does not change the permission profile, boundary expansions, or any other authority dimension.

/boundaries — Boundary Expansion management

Alias: /boundary
Requires approval: Yes
/boundaries lists, grants, and revokes Boundary Expansions. Because grant and revoke mutate authority, they require interactive approval. Without a TTY, the command fails closed.
/boundaries list
/boundaries grant <scope>
/boundaries revoke <expansionId>
The activeBoundaryExpansionCount in the AuthorityProjection reflects only non-revoked, non-expired expansions. Revoked or expired expansions are excluded from the count.
Granting a Boundary Expansion widens the authority scope for the current activation. The expansion is durable — it persists until explicitly revoked or until it expires. Always review the scope before granting.

Build docs developers (and LLMs) love