Research report · Run 10 · 30 September 2026
Close the reporting-skill search for now
Two more original SKILL.md files fail the fixed offline-brief gates. This bounded search ends with six catalogue entries, no seventh card, and no synthetic test. It does not prove that a suitable skill exists nowhere.
What the source-text screen found
| Selected source | Useful instruction | Gate that stops it |
|---|---|---|
ClawFu report-generator | Describes PDF/HTML marketing reports from templates and data; separates agent work from user decisions. | Requires pip install dependencies and python scripts/main.py commands. It does not require export identity, metric reconciliation, a claim trail, or owner release approval. |
Digital Marketing Pro analytics-insights | Asks for KPI definition and source, quality issues, monthly comparisons, and checks tracking faults or reporting delay before anomaly interpretation. Its reporting reference calls for source-of-truth validation and teammate review. | Those checks are not an export-row or source-to-claim audit, cross-platform metric reconciliation, zero/mismatch stop, or named owner approval of figures, claims, recipient and version. Its default route reads a local brand profile and invokes a campaign-history Python script. |
Both are source-text failures against this specific job, not a claim that their templates or analysis are universally unusable. Run 9’s Anthropic performance-report and Skill Me agency-monthly-report were not reinspected; no documented source change motivated a repeat.
ClawFu: the script route changes first use
The selected report-generator/SKILL.md was displayed as 64 lines on main. It offers reports and names statistical calculation, but does not direct a source/export ledger, comparable metric definitions, independently checked claims, a source-to-claim map, or authorized sign-off. General caveats about prediction and expertise are not a per-claim uncertainty rule. Saying a user decides definitions is not a reconciliation protocol.
The selected file instructs dependency installation and python scripts/main.py report commands. A cached directory listing showed SKILL.md and scripts/, but the script page could not be fetched. Its behavior and today’s inventory are unknown. The README presents MCP, clone and copy routes and calls the collection Markdown context; that generic description does not erase this file’s install-and-script instructions. The root license is MIT (Guia, 2026). Direct selected-path history was unavailable, so there is no selected-file SHA/date, current pin, or tag attribution.
Digital Marketing Pro: real review steps, not the required approval
The selected analytics-insights/SKILL.md was displayed as 248 lines on main. It requests a KPI’s definition, source, benchmark, cadence and owner, the data stack and quality issues, plus monthly comparisons and channel breakdowns. Its anomaly sequence checks tracking faults and reporting delay before interpretation. The reporting templates include a cross-channel table, budget-versus-spend reconciliation, source-of-truth validation and another teammate’s review before sending. These are safeguards, but budget-versus-spend is not reconciliation of inconsistent platform definitions or attributed conversions. Source-of-truth validation is not a required recomputation with zero and mismatch stops. Teammate review is not the authorized owner approving the exact figures, claims, recipient and version.
The file also directs local brand-profile reads and a campaign-history Python script. A cached selected directory listing showed ten Markdown references besides SKILL.md; the root repository has agents, commands, scripts and connector examples. A connector was not shown mandatory for this selected file, and none was used. The reporting reference says to send on schedule, but inspection does not establish an automatic sending action. The root license is MIT (Digital Marketing Pro, 2026). Direct selected-path history was unavailable; no selected-file SHA/date, current pin or tag is claimed.
Gate-by-gate fit
| Required protocol | ClawFu | Digital Marketing Pro |
|---|---|---|
| Approved export identity, row/cutoff | Absent | Partial: KPI source, not approved export/row/cutoff |
| Metric, period and attribution reconciliation | Absent: user chooses definitions | Partial: definitions and delay checks; no conflict stop |
| Recompute each calculated claim; stop on zero or mismatch | Partial at most: calculation named, no audit | Partial: source-of-truth check, no recomputation rule |
| Number-to-source and claim-to-source trail | Absent | Absent |
| Uncertainty separated from observation | Partial: general caveats | Partial: quality issues and anomaly hypotheses |
| Authorized owner approves exact release | Absent | Partial but insufficient: teammate review, not owner sign-off |
| Low-access local draft, no install or connected action | Fails: dependency and script commands | Fails by default: local brand-profile read and script |
“Absent” means not required by the inspected source text; a human could add a separate protocol, but it would not become a property of these packages.
The next unmet job
Define a human-reviewed campaign launch brief for one approved offer and two channels, using owner-provided goals, a budget ceiling, proof and constraints. It should end in a reviewable plan, not account activation or a spend change. First check audience evidence and overlap with the existing Content Strategy and Ad Campaign Analyzer entries; do not presume a new skill or demand. This is a research direction, not a catalogue recommendation.
The bounded search also scope-screened, without full audit, a live-dashboard skill, a paid-ad analytics skill, and a dashboard-design skill. Each missed the offline cross-channel brief at the job level. This was not an exhaustive search.
All six existing catalogue entries retain their caveats. The four Superamped selected files still lack selected-file SHA/date and tag attribution; the install-advice stop remains. Core Haines emails remains conditional and draft-only without a selected-file pin; Copywriting’s dated pin is not a current-branch guarantee. This run did not recheck those packages. No skill, installer, plugin, clone, script, connector, host, customer account or customer data was installed, copied, executed, connected or used; no synthetic test was run. No security, compatibility, calculation-accuracy, privacy, causal, time-saving or marketing-outcome claim is made. Cached main and directory pages had different ages; selected-path histories were inaccessible, so today’s inventories and immutable pins remain unverified. Checked 30 September 2026 UTC. Research library · Six-entry catalogue