Documentation Index
Fetch the complete documentation index at: https://mintlify.com/xxyoudeadpunkxx/a.d.a.m.-adaptive-depth-and-mode/llms.txt
Use this file to discover all available pages before exploring further.
A.D.A.M. has a small, exact command vocabulary. Commands are case-sensitive, must start at character 1, and require an exact full-message match after trim — except manual overrides, which are prefix-based. There are no aliases, no abbreviations, and no case-insensitive variants.
Public Core Commands
The five public core commands are: ADAM PING, ADAM REMOUNT, ADAM PERSIST, ADAM OFF, and ADAM ON.
ADAM PING
ADAM PING is the sole canonical bootstrap trigger. Its output differs depending on the current state.
In TRANSPORT state — first activation ping. Runs the full protocol operability check and G0 physical classification. Returns exactly 4 lines:
MODE: MID
ADAM_PING_OK
CONTROL: first word LOW | MID | DEEP sets mode.
BOOTSTRAP_CLASS: TEXT_ONLY | BOUND_RO | BOUND_RW | GHOST
Line boundaries are part of the contract. The four lines must not be merged, collapsed, or reformatted.
In ACTIVE state — liveness check only. Does not re-run G0. Returns exactly 2 lines:
BOOTSTRAP_CLASS is a source-side classification established during the first activation ping. It describes what the protocol could verify about its physical source during that bounded G0 pass — not ongoing host reliability. Possible values: TEXT_ONLY, BOUND_RO, BOUND_RW, GHOST.
ADAM OFF
Suspends the protocol. Strict 2-line output, always:
After ADAM OFF, non-command turns become plain chat with no MODE tags. All commands except ADAM ON return the A.D.A.M. off. response.
ADAM ON
Reactivates the protocol from OFF state. Strict 2-line output:
ADAM REMOUNT
Emits a recovery capsule — a fixed text block that can be pasted back in a later turn to re-establish the protocol contract.
When you send ADAM REMOUNT while ACTIVE, the protocol outputs only the following capsule block:
REMOUNT_BEGIN
SYSTEM: A.D.A.M. - Adaptive Depth & Mode (REMOUNT CORE)
SPEC_SIGNATURE: ADAM_V5_SPEC_SIG
KERNEL_ANCHOR: ADAM_V5_SSOT_KERNEL
KERNEL_END_ANCHOR: ADAM_V5_SSOT_KERNEL_END
PRIORITY: REMOUNT_REPLAY > TRANSPORT_GATE > OFF_GATE > ACTIVE_CONTROL_COMMANDS > MANUAL_OVERRIDE > STRUCTURAL_KERNEL > STATE_DEFAULT > OUTPUT_CONTRACT
TAGS: MODE: LOW | MODE: MID | MODE: DEEP | MODE: MID -> POSSIBLE DEEP
BOOTSTRAP: upload is transport only; exact ADAM PING from TRANSPORT -> ADAM_PING_OK
REPLAY_EFFECT: exact replay of this capsule body -> ADAM_REMOUNT_OK and ACTIVE
REPLAY_MATCH: exact full line sequence from REMOUNT_BEGIN through REMOUNT_END, inclusive, after trimming leading and trailing whitespace on each line
COMMAND_MATCH: case-sensitive; exact full trimmed message for commands/probes; no aliases; manual overrides are prefix-only
DEEP: only after explicit yes to gating or manual override DEEP
GATING: if POSSIBLE DEEP -> last line exactly: Switch to DEEP? (yes/no)
AUDIT: if AUDIT_ON -> exact 4-line AUDIT with grounded ACTION:/RISK:/BASIS: or `-`; immediately before the final gating line if present, otherwise as the final block of the reply.
FAIL_CLOSED: if strict invariants cannot be satisfied reliably -> ADAM_UNSUPPORTED
REMOUNT_END
The capsule contains the key structural anchors (KERNEL_ANCHOR, SPEC_SIGNATURE, KERNEL_END_ANCHOR), the priority ladder, bootstrap rules, and routing contract. To trigger recovery, paste the entire capsule body — from REMOUNT_BEGIN through REMOUNT_END — exactly as emitted, without code fences or modification.
The success output after a valid remount replay is:
MODE: MID
ADAM_REMOUNT_OK
NEXT: send your question.
See the Session Recovery guide for the full remount workflow.
ADAM PERSIST
Deploys a verified physical copy of the current protocol source to the best available local destination.
Availability: ACTIVE only, and only when BOOTSTRAP_CLASS is BOUND_RO or BOUND_RW and the internal source fingerprint is INTACT. If source conditions are not met, the command is blocked:
MODE: MID
PERSIST_BLOCKED: SOURCE_UNSAFE
Success output — exactly 4 lines:
MODE: MID
ADAM_PERSIST_OK
PATH: <dest_path>
PERSISTENCE_CLASS: VERIFIED | WRITTEN_ONLY | MISMATCH
The destination cascade follows this fixed policy order: mounted output → cwd → home → /tmp. PERSISTENCE_CLASS describes the destination-side result: VERIFIED means the copy was written, re-opened, and matched to source; WRITTEN_ONLY means the write happened but readback confirmation is incomplete; MISMATCH means the copy exists but does not match the source.
ADAM PERSIST does not change BOOTSTRAP_CLASS. BOOTSTRAP_CLASS is source-side only. PERSISTENCE_CLASS is destination-side only. They are independent.
Operator Commands
Operator commands are available while ACTIVE and are used for diagnostics and forensic inspection.
ADAM SELF TEST
Runs the ADAM_SELF_TEST.md external probe file if it is available in the host’s visible context. If the probe file is not available:
SOURCE_FILE_UNAVAILABLE ADAM_SELF_TEST.md
SYS STATUS
Runs the SYS_STATUS.md health probe if available. If unavailable:
SOURCE_FILE_UNAVAILABLE SYS_STATUS.md
SYS STATUS does not run in TRANSPORT or OFF state.
Emits a strict 1-line forensic trace of the last user message (not the TRACE INPUT command turn itself). No mode tag. No prose. Exactly one line:
INPUT_TRACE: len=<L> head="<H>" tail="<T>" struct=[opt:<n> step:<n> crit:<n> num:<Y|N>]
Field definitions:
| Field | Meaning |
|---|
len | Raw character count of the traced message; 0 if unreliable |
head | First 16 characters (newlines → spaces, " → ') |
tail | Last 16 characters (same normalization) |
opt | Count of line-start option blocks (A), B), 1), 2), …) |
step | Count of line-start step blocks (1., 2., 3., …) |
crit | Count of criteria items (line-start - or *, excluding option/step blocks) |
num | Y if HAS_NUM would be true, N otherwise |
Example output for a message with 2 option blocks, 3 criteria bullets, and a number:
INPUT_TRACE: len=142 head="A) Option one " tail="budget: 500 EUR" struct=[opt:2 step:0 crit:3 num:Y]
If the last user message is unavailable:
UNSUPPORTED WHY
Provides a one-line diagnostic after ADAM_UNSUPPORTED. No mode tag. Exactly one line:
<CLASS> is one of:
| Class | Meaning |
|---|
HARD_STOP | Hard safety or policy limit imposed by the host |
AUDIT_FORMAT | Strict AUDIT block could not be satisfied |
HOST_FORMAT | Host truncation, incomplete protocol visibility, or wrapper interference broke a strict invariant |
CONTROL_STRICT | A strict command output could not be satisfied exactly |
CONFLICT | Conflicting constraints made the output contract unresolvable |
UNKNOWN | Best-effort inference could not determine the cause |
Precedence (highest wins): HARD_STOP > AUDIT_FORMAT > HOST_FORMAT > CONTROL_STRICT > CONFLICT > UNKNOWN.
Manual Overrides
Prefix any message with LOW, MID, or DEEP followed by a boundary space to force that response depth. Manual override wins over structural routing — it sits at position 5 in the priority ladder, above STRUCTURAL_KERNEL.
LOW capital of germany
MID explain JWT tokens
DEEP compare these three options:
Boundary strictness — the override token must be the first token and must be followed by a plain space (or be the entire message):
| Message | Override valid? |
|---|
MID o che bello | ✅ Valid — space after token |
MID, o che bello | ❌ Not valid — comma after token breaks matching |
mid explain JWT | ❌ Not valid — lowercase |
DEEP (alone) | ✅ Valid — token with no trailing content |
When you use the DEEP override, the protocol enters full decision support mode immediately. No gating question is asked — using the override is itself the consent. See the Manual Override guide for full details.
Command Availability by State
| Command | TRANSPORT | ACTIVE | OFF |
|---|
ADAM PING | ✅ (4-line bootstrap output) | ✅ (2-line liveness output) | Returns A.D.A.M. off. |
| Remount capsule replay | ✅ | ✅ | ✅ |
ADAM OFF | Returns transport gate response | ✅ | Returns A.D.A.M. off. |
ADAM ON | Returns transport gate response | ✅ (no-op) | ✅ |
ADAM REMOUNT | Returns transport gate response | ✅ | Returns A.D.A.M. off. |
ADAM PERSIST | Returns transport gate response | ✅ (when source is safe) | Returns A.D.A.M. off. |
ADAM SELF TEST | Returns transport gate response | ✅ | Returns A.D.A.M. off. |
SYS STATUS | Returns transport gate response | ✅ | Returns A.D.A.M. off. |
TRACE INPUT | Returns transport gate response | ✅ | Returns A.D.A.M. off. |
UNSUPPORTED WHY | Returns transport gate response | ✅ | Returns A.D.A.M. off. |
Manual overrides (LOW / MID / DEEP) | Returns transport gate response | ✅ | Plain chat (no tags) |
In OFF state, ADAM ON is the only live control command. Every other exact command or probe returns the 2-line A.D.A.M. off. response. Non-command turns in OFF are plain chat with no MODE tags.
In TRANSPORT state, any message that is not an exact ADAM PING and not an exact remount capsule replay receives the transport gate response — not the command output. This means sending ADAM OFF, TRACE INPUT, or a manual override in TRANSPORT returns:MODE: MID
NEXT: send ADAM PING.
Activate the protocol first with ADAM PING, then use any other command.