AgentAbuse Working draft

Documentation

Glossary

The vocabulary of the report.

Core

abuse report
A signed, per-reporter hash-chained accusation citing existing evidence by hash. Proves an accountable accusation was made; never adjudicates it. Reference implementation: sm-abuse-report.
report
The eleven-field signed JSON object: protocol, reporter_id, accused_id, accused_role, category, evidence_refs, narrative, prev_hash, issued_at, sig, pubkey. See What is an Abuse Report.
accused_role
Which side stands accused: agent or principal. The field that makes the schema bidirectional without needing two report types.
category
A closed, versioned vocabulary naming the kind of harm, or other — where narrative carries the substantive claim. See the taxonomy on What is an Abuse Report.
evidence_refs
A non-empty array of {type, hash} pointers to existing sm-arp receipts or sm-aae envelopes — never a bundled copy of the artifact. A report may cite more than one.
narrative
The reporter's required free-text account. Load-bearing when category is other.
prev_hash
The hex SHA-256 of this reporter's previous report, or null for its first report. What chains one accusation to the next.
report hash
The lowercase-hex SHA-256 of the full signed report (including sig) — the value a successor's prev_hash points at.
chain
A reporter's reports linked genesis-first by prev_hash. Reconstructed by order_chain, which returns the ordering or rejects a set that isn't exactly one intact chain.
canonicalization
The deterministic serialization the signature covers: sorted-key, compact-separator JSON over the report with sig removed, UTF-8 encoded. Identical inputs always produce byte-identical canonical bytes.
gap / fork / foreign-chain splice
The three chain defects order_chain rejects: a missing interior report, two reports claiming the same predecessor, and a prev_hash that reaches into another reporter's chain, respectively.
confirmed / hash_mismatch / type_mismatch / not_party
The four possible results of verify_evidence_citation, run by a verifier who holds the cited artifact: the citation checks out, the hash doesn't match, the declared type doesn't match, or the reporter isn't actually a party to the artifact.
unconfirmed
The state of a citation nobody has checked because the verifier doesn't hold the cited artifact — distinct from both confirmed and any failure result. Not a bug; an honest reflection of what wasn't checkable.

Stack acronyms

Canonical expansions used across the Stellarminds agent-evidence stack.

ARP
Agency Receipt Protocol — the signed receipt envelope: what an agent did. See Primitives.
AAE
Attested Action Envelope — the signed, per-agent-chained record of a pre-action authorization verdict.
DAT
Delegated Authority Token — the principal-signed grant defining what an agent was allowed to do.
REP
Portable Reputation Credential — portable reputation computed over an agent's ARP receipt history (package sm-rep). A named, not-yet-wired future direction for abuse reports.