A mandate that names a dh concept states only the name, not the method — the orchestrator sees the outcome (a verdict, a gate, a report) and can misinfer the mechanism from watching what it does. Load this skill instead of inferring; each entry states the method, not just the outcome, and points to the file that is authoritative if this entry ever drifts from it.
Terms
RT-ICA (Reverse Thinking - Information Completeness Assessment) — works backward from the
stated goal through its prerequisite chain: what must exist to reach the goal, then what each of
those things requires, recursively back to the current state. Classifies each prerequisite as
AVAILABLE, DERIVABLE, or MISSING. dh:rt-ica blocks planning on any MISSING condition (S2
implementation gate); dh:planner-rt-ica is the non-blocking sister used during grooming, which
localizes a MISSING condition to the affected task instead of halting. Canonical: load the
dh:rt-ica skill.
ARL (Autonomous Refinement Loop) — research into the prerequisites for autonomous agent
execution without synchronous human blocking gates: failure categories, and the conditions under
which a machine-verifiable check can replace human judgment. dh's human-touchpoint model is
ARL-derived — see the S1-S7 entry below. Canonical: load the plugin-creator:arl skill.
SAM (Stateless Agent Methodology) — the 7-stage development pipeline this harness implements
(S1-S7, below). "Stateless" means each stage's state lives in a provider-owned artifact, never in
conversation history — a stage reads its input from the prior stage's registered artifact, not
from what an agent remembers saying earlier. Canonical: load the dh:dh-meta-docs skill.
S1-S7 — the SAM pipeline stage names, in order: S1 Discovery (understand the feature,
codebase, constraints) → S2 Planning + RT-ICA (generate a plan, gated by dh:rt-ica) → S3
Context Integration (validate the plan against actual codebase state) → S4 Task
Decomposition (break the plan into executable tasks) → S5 Execution (implement tasks via
language-specific specialists) → S6 Forensic Review (verify each task against its acceptance
criteria) → S7 Final Verification (certify the feature meets the original requirements). Not
every stage requires human review — ARL-derived constraint analysis decides when to escalate, not
a fixed checkpoint per stage. Canonical: load the dh:dh-meta-docs skill.
Impact Radius — the backlog-item section whose Systems Inventory records the estimated impact
set across software, data, runtime, prompts, controls, people, and processes, with causal paths,
risk, verification, exclusions, and owned unknowns. It is written by @dh:impact-analyst before
planning; the feasibility and staleness gates consume it. Older items may hold a file-centric
version under Impact Radius or Resources; consumers retain that fallback. Canonical:
plugins/development-harness/agents/impact-analyst.md.