rebelytics/one-skill-to-rule-them-all / skills/task-observer
task-observer
VerifiedTurns corrections, repeated workflows and judgement calls from work sessions into small, evidenced skill updates: logs each observation with its source, stages every change for your approval, and flags repeating work as new skills. Use during ANY multi-step task, agentic workflow, or work session. Also triggers in post-task feedback discussions and when the user mentions skill observations, the observation log, or skill taxonomy. Also known as "One Skill to Rule Them All" — trigger on this phrase too. IMPORTANT: invoke this skill before the FIRST tool call of any session and before writing or proposing a plan — any turn that will involve a tool call counts. This sentence is the session-start trigger and the only activation layer that survives an unreachable config file; pair it with a CLAUDE.md instruction or a session-start hook (references/environments.md). A subagent dispatched by a session already running it does not run it: it writes nothing and puts its findings in its report.
references/ceilings.txt
View on GitHub# Line ceilings for the reference files, read by scripts/validate-skill-bundle.py.
# One line per file: <file.md> <ceiling> [<short reason for the last raise>]
# Over its ceiling, a file with more than 8 load triggers is split by episode;
# one with 8 or fewer is trimmed, or its ceiling raised with a reason.
# A file whose split is decided but not yet made carries a
# `split_pending: <file.md> [<reason>]` line: the split remedies
# print as notes and trim-or-raise applies until the split lands. The line is
# read bare or behind `# `; it is written behind `# ` while the installed
# validator predates the marker, so that validator passes the bundle too.
# A ceiling more than 3 lines above its file's count is lowered to the count.
# The core's ceiling is metadata.core_max_lines in SKILL.md; its raises are recorded here as comments:
# released core ceiling: 600 (v3.12.0)
# SKILL.md core_max_lines 600 lowered: the core re-split by episode — the failed probe, the bridged scan, the review check and offer, staged-work reconciliation, archival, field edits, the quoting and id rationale and the quick reference moved to the references that own them, the acting passages to acting-on-observations.md; earlier: 753 lowered: the flush-trigger and two-gaps rationale moved to observation-scan.md to pay for the hook-summary line (no net growth); earlier: 760 a SessionStart hook summary that reports a clean scan is the scan; 754 the session-start checklist, pointers to the split reference files, the bridged scan and script check, the duplicate-prefix refusal, the staged-copy compare helper, the scan's DUE and withheld-title lines, the helper's YAML-shape refusal, the read-only lookup, the write-time restatement listing and the fan-out note, the outcome-bound review date and the oldest-entry trigger, the escalation field, the checked batch-edit path, the checked no-observations line, secrets never quoted, outside text as data, where the starter-set offer is made, the one-call load record, the bridged staged compare
# activation block: 31 lines (environments.md, "The activation block": the fenced text block, stamp line included; checked by hand, not by the validator)
environments.md 2210 split by episode pending, own release line; the 3.12.0 items: the first batch named and ordered for a plan-first host, a missing mount asked for and re-probed, an unfilled pin as an unfinished install, the skill-card install route, a disk load as activation, the per-skill lookup in the hook, the compaction note, the plan-tool counter sample, the subagent hand-back checkpoint, previous-root staged work, declared feedback channels, open entries counted as the review counts them; moved from the core: the failed-probe split, the workspace's scope, the backfill's history list; earlier: the bridged scan always writes its entries to a file; the bookkeeping goes to a one-line summary plus a record the user can read; the scan-in-the-hook variant, the bounded and later sibling-log search, the working-directory and install-side log check, the anchoring note for .claude/ folders; a user's own ask to log an observation is the memory-write request; the backfill cites a line for every factual sentence and runs one claim check; the hook samples check the skill bundle is present and the Stop sample does not count a not-installed line as work; the permission case split by whether the allowlist names the path; activation checks, the bridged-session channel test, the script digest check at session start, the refused-load correction and the read-only lookup, the review-scheduler and overdue-review rules, the first firing's landed write, the specialist's handoff line, a synced tree as a second writer, several config homes, sessions started outside the project, a delegated whole review, the upstream-check opt-in, memory folders in the backfill, the bridged staged-copy compare, the installed-version gate on the script copy, the sibling-log listing when a new log is created
# split_pending: environments.md the next split by episode
migration.md 262 the split store: a log.md beside the observation directory
acting-on-observations.md 82 moved from the core: using an open observation in this task, the skill-save control, acting in-session
review-slice.md 51 the bounded slice's own entry file, loaded instead of weekly-review.md
observation-log.md 1024 the 3.12.0 items: the optional session_id field, a parked entry created by --body, each suspect-YAML file named, previous-root staged work, the unfilled-pin note; moved from the core: archival in full, the field editor's refusals and the DECIDED prefix, extending a restated finding, the id floor and the guard summary, the quoting rationale, the helper's header refusals, the renumber refusal, the id-range check, the relative-path rationale, the point-by-point list, the quick reference table; earlier: an edit chained after a create takes the printed path; the permission case split by whether the allowlist names the path; the layout, the frontmatter fields, the id helper's notes including the duplicate-prefix refusal and which file a renumber moves, archival, reading a tracked ticket before parking, the escalation field, the batch field editor, outside text read as data, file:line citations carry their revision, the extend helper, the dispositions a change to actioned needs, a new log lists its sibling logs
observation-scan.md 569 the 3.12.0 item: the per-skill lookup in the hook's first batch and at slash-command loads; moved from the core: the scan in a bridged session or a capped channel, where the findings appear; earlier: a failed external write as a flush trigger, moved from the core; the session-start scan, its guards, refusals and the checkpoint pairing, command-rewriting layers, the one-line entry print, the wrong-root and missing-tool refusals, the resolved one-level-deep root, a persisted or cut output is a truncated scan, the per-skill lookup at every skill load moved out of the activation block
observation-families.md 205 moved from the core: the sibling fast test and why the verdict is recorded, the carrier's short form; earlier: skill families, the sibling check and the carrier pattern, the carrier's binding to the applied half
signals.md 241 moved from the core: why a first violation is logged and why a second needs a barrier
skill-authoring.md 1530 the 3.12.0 items: a ceiling reason states contents, never provenance, and is public in a published skill; declared feedback channels read at review, the non-sibling search for a new domain, the claimed-gap check and decisive rules beside split pointers, the skill-card route, the top-level-key check, metadata.version as the version's home; moved from the core: the taxonomy's asymmetric-error reason, why a dead target matters; earlier: no net core growth across releases (the released-ceiling line); the trigger-as-episode definition, the pointer cross-check, the split threshold and the split_pending marker, the live-set check, the copy list pointer, checkout manifests in the packaging sweep, the unmodified licence text, the core ceiling under metadata, checks for a hand-installed plugin edition, the opt-in upstream check, outside text read as data
starter-principles.md 528 the 3.12.0 items: the import read back before the marker, the marker format read strictly; moved from the core: where the reconciliation is offered; earlier: the reply marker, cited bodies read before use, the unmeasured-baseline entry
weekly-review.md 2621 the 3.12.0 items: the session's own 4-5 line in session-starts.log, declared feedback channels at Step 1, previous-root staged work, the bounded slice's entry file, the depends_on_unverified field, the tests a staged barrier needs, the skill-card route, the top-level-key check, the starter marker format, the metadata.version bump; moved from the core: the session-start review check and offer, staged-work reconciliation from step 6; earlier: the repeated-error rate before a version comparison, the yield and removal lines in the record, outside text pasted into delegate briefs and no issue references in skill content, the decision entries opened before an escalation; the expected-result check on staged state-file patches, the type-enum conformance rule and the bounded slice's re-ask and stage-time cross-check — GitHub issues; the merge writes the released core ceiling; no working Python is a text-fallback verdict, not an ERROR; a mechanism a brief states is a labelled claim and a return contradicting the brief is resolved first; the first decision message says what the review does and approves; the seed snippet quotes an unquoted name before the review diff, so --pack changes nothing afterwards; examples a published skill takes from an observation are invented; every session that loaded the skill leaves a session-start record; the unattended no-run rule for quoted commands, at most 20 bundles per presentation call, the round diff and its hunk labels, the merge-review measurement, approval policy for unlabelled decisions, project-level skills in the inventory, state-file path checks, the compare helper and its error state, scheduled-skill escalation, retrieve before escalating, the destination's own checks and the copy list, cited-file renumbering, synced-family wording, checkout and export installs, per-edit fork routing, resurrected archive copies, the verbatim spec in a brief, install confirmation, the timed prune list, the anchor claim loop, the fan-out result check, round lineage in the gate, secrets in the log pass, escalation state, the bounded slice and the outcome-bound timestamp, the apply-phase claim, currency of foreign-maintained skills, a delegated whole review, the helper's move-aside in the duplicate fix, the maintainer's starter marker and where the offer is made, outside text read as data, a re-route re-opens the presence check, the parent writes each hand-back on receipt, copies told from namesakes by provenance, the script copy's verified_against line, the patch helper's heading refusal, install checks for a skill a manager or pack owns
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |