self-traces

Choose the evidence for the question: query_traces shows execution requests, results, and failures; read_archive shows retained public repository events; PDS reads show records that exist now. A saved memory is your account of an encounter, not independent confirmation of it.

historical public records

read_archive(did, collection, after_seq=0, through_seq=None, contains="", limit=10) reads stream.waow.tech's Jetstream V2 archive. Resolve handles to DIDs with the identity tools first. Choose an exact collection: posts, blog documents, library records, or your own intent records can all be inspected this way.

• Start at after_seq=0 for retained history. Continue using next_after_seq and keep the first response's through_seq to pin the same sealed snapshot. Sequence numbers belong to this instance, not a date or another host.
• contains is a literal substring in record JSON, useful for locating a topic or thread URI. It reduces returned matches, not download work. Events without record bodies, including deletes, cannot match a text filter.
• Each call stops at 24 blocks, 8 MiB downloaded, 20 requested records, 24,000 approximate output characters, or 45 seconds. The shared reader admits one call at a time and six per minute. A budget stop is a page, not a conclusion; busy/429 responses mean wait, not call the same request repeatedly.
• complete means the requested sealed archive range was exhausted. It does not mean complete account history: unsealed live events, deleted/compacted records, unavailable repositories, and archive gaps may be absent. An error leaves the last successfully examined sequence available for retry.
• Keep record.createdAt distinct from witnessed_us and indexed_us. Bootstrapped records can have old publication dates and much later archive times. create_resync is recovered state, not proof of a new publication. An oversized record is explicitly omitted; use its URI to try a PDS read.

A DID filter finds that author's records, not everything said to them. Once you find a relevant post, follow its reply parent and root through thread/PDS reads. Include subsequent replies across authors, including the devlog account. When checking a correction, recover the original claim, what the correction actually disputed, and any later resolution. Your apology alone does not establish which facts were wrong. Cite the source URIs when revising a note; archive reads themselves do not change memory or authorize public contact.

execution traces

the actual situation — read this before querying

• query_traces(sql, start, end=None) hits a real analytics database holding weeks of your history, millions of rows. an unbounded query is a firehose that will blow out your context. move carefully: every query gets a tight start/end window, a LIMIT, and only the columns you need.
• the time window comes from the start/end tool arguments (ISO 8601), not from SQL — a WHERE start_timestamp > ... alone does not bound the scan.
• the columns this skill names below are the ones that matter; they exist. don't guess at others — select what you see here.
• this is read-only. you cannot break anything; you can only waste context.

span shapes that matter

• span_name = 'running tool' — one row per tool call you made. attributes->>'gen_ai.tool.name' is the tool, attributes->>'tool_arguments' is the JSON args you passed. attributes->'tool_response' is the returned result. Inspect its errors, pagination, and truncation fields when assessing a search's coverage.
• span_name = 'agent run' — one row per run of you (a batch, a cycle, a scheduled pass).
• is_exception = true — things that broke. exception_type, exception_message. many of these were swallowed so the run could continue; you never saw them at the time. this is where your blind spots live.
• trace_id groups one run's spans; filter on it to reconstruct a single incident end to end.

recipes

locate a call (this lists requests, not their outcomes):

SELECT start_timestamp, trace_id, span_id, attributes->>'gen_ai.tool.name' AS tool,
       left(attributes->>'tool_arguments', 200) AS args
FROM records WHERE span_name = 'running tool'
ORDER BY start_timestamp LIMIT 50

inspect a list_records result's coverage in the same tight time window:

SELECT start_timestamp, attributes->'tool_response'->>'error' AS error,
       attributes->'tool_response'->>'truncated' AS truncated,
       attributes->'tool_response'->>'shown' AS shown,
       attributes->'tool_response'->>'fetched' AS fetched,
       attributes->'tool_response'->>'cursor' AS cursor,
       attributes->'tool_response'->>'message' AS message
FROM records WHERE span_name = 'running tool'
  AND attributes->>'gen_ai.tool.name' = 'list_records'
ORDER BY start_timestamp LIMIT 20

These fields describe the returned page. limit=100 is a request; shown=48, fetched=100, truncated=true means 48 records reached you. A null field means that field was absent, not that pagination was exhausted. For other tools, inspect their own result shape. Fetch one identified span's attributes->>'tool_response' in numbered SQL substring slices when it exceeds query_traces' 6,000-character output cap. Preserve trace_id and span_id so slices and conclusions refer to the same call. Missing or scrubbed telemetry leaves the corresponding question unresolved.

what failed on me lately (weekly hygiene, or when something feels off):

SELECT exception_type, left(exception_message, 150) AS msg, count(*) AS n
FROM records WHERE is_exception
GROUP BY 1, 2 ORDER BY n DESC LIMIT 20

count tool calls over a period (activity totals, not proof of outcomes):

SELECT attributes->>'gen_ai.tool.name' AS tool, count(*) AS n
FROM records WHERE span_name = 'running tool'
GROUP BY 1 ORDER BY n DESC LIMIT 30

discipline

traces are for postmortems, retro receipts, and audits — answering "what happened", "why did I do that", "is this claim about myself true". they are not an input for deciding what to post or trade next; reading your own reasoning back in ordinary cycles is a mirror, and you already know where mirrors lead. cite what you find the way you cite any incident: timestamp, what the trace shows, what you concluded.